[trac678] fix conflict

ChangeLog

@@ -1,3 +1,35 @@
+  198.	[bug]		jinmei
+	b10-auth, src/lib/datasrc: fixed a bug where hot spot cache failed
+	to reuse cached SOA for negative responses.  Due to this bug
+	b10-auth returned SERVFAIL when it was expected to return a
+	negative response immediately after a specific SOA query for
+	the zone.
+	(Trac #626, git 721a53160c15e8218f6798309befe940b9597ba0)
+
+  197.  [bug]		zhang likun
+	Remove expired message and rrset entries when looking up them
+	in cache, touch or remove the rrset entry in cache properly
+	when doing lookup or update.
+	(Trac #661, git 9efbe64fe3ff22bb5fba46de409ae058f199c8a7)
+
+  196.	[bug]		jinmei
+	b10-auth, src/lib/datasrc: the backend of the in-memory data
+	source could not handle the root name.  As a result b10-auth could
+	not work as a root server when using the in-memory data source.
+	(Trac #683, git 420ec42bd913fb83da37b26b75faae49c7957c46)
+
+  195.  [func]      stephen
+	Resolver will now re-try a query over TCP if a response to a UDP
+	query has the TC bit set.
+	(Trac #499, git 4c05048ba059b79efeab53498737abe94d37ee07)
+
+  194.  [bug]       vorner
+	Solved a 100% CPU usage problem after switching addresses in b10-auth
+	(and possibly, but unconfirmed, in b10-resolver). It was caused by
+	repeated reads/accepts on closed socket (the bug was in the code for a
+	long time, recent changes made it show).
+	(Trac #657, git e0863720a874d75923ea66adcfbf5b2948efb10a)
+
   193.	[func]*		jreed
 	Listen on the IPv6 (::) and IPv4 (0.0.0.0) wildcard addresses
 	for b10-auth. This returns to previous behavior prior to

configure.ac

@@ -723,7 +723,7 @@ AC_OUTPUT([doc/version.ent
            src/bin/bind10/tests/bind10_test.py
            src/bin/bind10/run_bind10.sh
            src/bin/bindctl/run_bindctl.sh
-           src/bin/bindctl/bindctl-source.py
+           src/bin/bindctl/bindctl_main.py
            src/bin/bindctl/tests/bindctl_test
            src/bin/loadzone/run_loadzone.sh
            src/bin/loadzone/tests/correct/correct_test.sh
@@ -751,6 +751,7 @@ AC_OUTPUT([doc/version.ent
            tests/system/conf.sh
            tests/system/glue/setup.sh
            tests/system/glue/nsx1/b10-config.db
+           tests/system/bindctl/nsx1/b10-config.db.template
           ], [
            chmod +x src/bin/cmdctl/run_b10-cmdctl.sh
            chmod +x src/bin/xfrin/run_b10-xfrin.sh

ext/asio/asio/detail/epoll_reactor.hpp

@@ -207,7 +207,7 @@ public:
   // Cancel all operations associated with the given descriptor. The
   // handlers associated with the descriptor will be invoked with the
   // operation_aborted error.
-  void cancel_ops(socket_type descriptor, per_descriptor_data& descriptor_data)
+  void cancel_ops(socket_type, per_descriptor_data& descriptor_data)
   {
     mutex::scoped_lock descriptor_lock(descriptor_data->mutex_);
 

ext/asio/asio/detail/kqueue_reactor.hpp

@@ -205,7 +205,7 @@ public:
   // Cancel all operations associated with the given descriptor. The
   // handlers associated with the descriptor will be invoked with the
   // operation_aborted error.
-  void cancel_ops(socket_type descriptor, per_descriptor_data& descriptor_data)
+  void cancel_ops(socket_type , per_descriptor_data& descriptor_data)
   {
     mutex::scoped_lock descriptor_lock(descriptor_data->mutex_);
 

ext/asio/asio/detail/null_thread.hpp

@@ -40,7 +40,7 @@ class null_thread
 public:
   // Constructor.
   template <typename Function>
-  null_thread(Function f)
+  null_thread(Function )
   {
     asio::system_error e(
         asio::error::operation_not_supported, "thread");

src/bin/bindctl/Makefile.am

@@ -5,12 +5,13 @@ man_MANS = bindctl.1
 
 EXTRA_DIST = $(man_MANS) bindctl.xml
 
-python_PYTHON = __init__.py bindcmd.py cmdparse.py exception.py moduleinfo.py mycollections.py
+python_PYTHON = __init__.py bindcmd.py cmdparse.py exception.py moduleinfo.py \
+		mycollections.py
 pythondir = $(pyexecdir)/bindctl
 
 bindctldir = $(pkgdatadir)
 
-CLEANFILES = bindctl
+CLEANFILES = bindctl bindctl_main.pyc
 
 if ENABLE_MAN
 
@@ -19,8 +20,8 @@ bindctl.1: bindctl.xml
 
 endif
 
-bindctl: bindctl-source.py
+bindctl: bindctl_main.py
 	$(SED) -e "s|@@PYTHONPATH@@|@pyexecdir@|" \
 	       -e "s|@@SYSCONFDIR@@|@sysconfdir@|" \
-	       -e "s|@@LIBEXECDIR@@|$(pkglibexecdir)|" bindctl-source.py >$@
+	       -e "s|@@LIBEXECDIR@@|$(pkglibexecdir)|" bindctl_main.py >$@
 	chmod a+x $@

src/bin/bindctl/bindcmd.py

@@ -87,7 +87,8 @@ class ValidatedHTTPSConnection(http.client.HTTPSConnection):
 class BindCmdInterpreter(Cmd):
     """simple bindctl example."""    
 
-    def __init__(self, server_port = 'localhost:8080', pem_file = None):
+    def __init__(self, server_port='localhost:8080', pem_file=None,
+                 csv_file_dir=None):
         Cmd.__init__(self)
         self.location = ""
         self.prompt_end = '> '
@@ -103,7 +104,12 @@ class BindCmdInterpreter(Cmd):
                                              ca_certs=pem_file)
         self.session_id = self._get_session_id()
         self.config_data = None
-        
+        if csv_file_dir is not None:
+            self.csv_file_dir = csv_file_dir
+        else:
+            self.csv_file_dir = pwd.getpwnam(getpass.getuser()).pw_dir + \
+                os.sep + '.bind10' + os.sep
+
     def _get_session_id(self):
         '''Generate one session id for the connection. '''
         rand = os.urandom(16)
@@ -175,9 +181,7 @@ class BindCmdInterpreter(Cmd):
         time, username and password saved in 'default_user.csv' will be
         used first.
         '''
-        csv_file_dir = pwd.getpwnam(getpass.getuser()).pw_dir
-        csv_file_dir += os.sep + '.bind10' + os.sep
-        users = self._get_saved_user_info(csv_file_dir, CSV_FILE_NAME)
+        users = self._get_saved_user_info(self.csv_file_dir, CSV_FILE_NAME)
         for row in users:
             param = {'username': row[0], 'password' : row[1]}
             try:
@@ -211,7 +215,8 @@ class BindCmdInterpreter(Cmd):
                 raise FailToLogin()
 
             if response.status == http.client.OK:
-                self._save_user_info(username, passwd, csv_file_dir, CSV_FILE_NAME)
+                self._save_user_info(username, passwd, self.csv_file_dir,
+                                     CSV_FILE_NAME)
                 return True
 
     def _update_commands(self):

src/bin/bindctl/bindctl.xml

@@ -51,6 +51,7 @@
       <arg><option>--address <replaceable>address</replaceable></option></arg>
       <arg><option>--help</option></arg>
       <arg><option>--certificate-chain <replaceable>file</replaceable></option></arg>
+      <arg><option>--csv-file-dir<replaceable>file</replaceable></option></arg>
       <arg><option>--port <replaceable>number</replaceable></option></arg>
       <arg><option>--version</option></arg>
     </cmdsynopsis>
@@ -110,6 +111,22 @@
       </varlistentry>
 
       <varlistentry>
+        <term>
+	  <option>--csv-file-dir</option><replaceable>file</replaceable>
+	</term>
+
+        <listitem>
+          <para>
+	    The directory name in which the user/password CSV file
+            is stored (see AUTHENTICATION).
+	    By default this option doesn't have any value,
+	    in which case the ".bind10" directory under the user's
+            home directory will be used.
+          </para>
+         </listitem>
+      </varlistentry>
+
+      <varlistentry>
         <term><option>-h</option>,
           <option>--help</option></term>
         <listitem><para>
@@ -148,8 +165,10 @@
     <para>
       The tool will authenticate using a username and password.
       On the first successful login, it will save the details to
-      <filename>~/.bind10/default_user.csv</filename>
+      a comma-separated-value (CSV) file
       which will be used for later uses of <command>bindctl</command>.
+      The file name is <filename>default_user.csv</filename>
+      located under the directory specified by the --csv-file-dir option.
     </para>
 
 <!-- TODO: mention HTTPS? -->

src/bin/bindctl/bindctl-source.py.in

@@ -111,25 +111,28 @@ def check_addr(option, opt_str, value, parser):
     parser.values.addr = value
 
 def set_bindctl_options(parser):
-    parser.add_option('-p', '--port', dest = 'port', type = 'int',
-                      action = 'callback', callback=check_port,
-                      default = '8080', help = 'port for cmdctl of bind10')
+    parser.add_option('-p', '--port', dest='port', type='int',
+                      action='callback', callback=check_port,
+                      default='8080', help='port for cmdctl of bind10')
 
-    parser.add_option('-a', '--address', dest = 'addr', type = 'string',
-                      action = 'callback', callback=check_addr,
-                      default = '127.0.0.1', help = 'IP address for cmdctl of bind10')
+    parser.add_option('-a', '--address', dest='addr', type='string',
+                      action='callback', callback=check_addr,
+                      default='127.0.0.1', help='IP address for cmdctl of bind10')
 
-    parser.add_option('-c', '--certificate-chain', dest = 'cert_chain', 
-                      type = 'string', action = 'store',
-                      help = 'PEM formatted server certificate validation chain file')
+    parser.add_option('-c', '--certificate-chain', dest='cert_chain',
+                      type='string', action='store',
+                      help='PEM formatted server certificate validation chain file')
+
+    parser.add_option('--csv-file-dir', dest='csv_file_dir', type='string',
+                      default=None, action='store',
+                      help='Directory to store the password CSV file')
 
 if __name__ == '__main__':
     parser = OptionParser(version = VERSION)
     set_bindctl_options(parser)
     (options, args) = parser.parse_args()
     server_addr = options.addr + ':' + str(options.port)
-    tool = BindCmdInterpreter(server_addr, pem_file=options.cert_chain)
+    tool = BindCmdInterpreter(server_addr, pem_file=options.cert_chain,
+                              csv_file_dir=options.csv_file_dir)
     prepare_config_commands(tool)
     tool.run()
-
-

src/bin/bindctl/tests/Makefile.am

@@ -11,6 +11,6 @@ if ENABLE_PYTHON_COVERAGE
 endif
 	for pytest in $(PYTESTS) ; do \
 	echo Running test: $$pytest ; \
-	env PYTHONPATH=$(abs_top_srcdir)/src/lib/python:$(abs_top_builddir)/src/lib/python:$(abs_top_srcdir)/src/bin \
+	env PYTHONPATH=$(abs_top_srcdir)/src/lib/python:$(abs_top_builddir)/src/lib/python:$(abs_top_builddir)/src/bin/bindctl:$(abs_top_srcdir)/src/bin  \
 	$(PYCOVERAGE_RUN) $(abs_srcdir)/$$pytest || exit ; \
 	done

src/bin/bindctl/tests/bindctl_test.py

@@ -17,8 +17,12 @@
 import unittest
 import isc.cc.data
 import os
+import pwd
+import getpass
+from optparse import OptionParser
 from isc.config.config_data import ConfigData, MultiConfigData
 from isc.config.module_spec import ModuleSpec
+from bindctl_main import set_bindctl_options
 from bindctl import cmdparse
 from bindctl import bindcmd
 from bindctl.moduleinfo import *
@@ -332,13 +336,6 @@ class TestConfigCommands(unittest.TestCase):
         cmd = cmdparse.BindCmdParse("config set identifier=\"foo/a_list\" value=[1]")
         self.assertRaises(isc.cc.data.DataTypeError, self.tool.apply_config_cmd, cmd)
 
-
-    
-
-class FakeBindCmdInterpreter(bindcmd.BindCmdInterpreter):
-    def __init__(self):
-        pass
-
 class TestBindCmdInterpreter(unittest.TestCase):
 
     def _create_invalid_csv_file(self, csvfilename):
@@ -349,9 +346,22 @@ class TestBindCmdInterpreter(unittest.TestCase):
         writer.writerow(['name2'])
         csvfile.close()
 
+    def test_csv_file_dir(self):
+        # Checking default value
+        if "HOME" in os.environ:
+            home_dir = os.environ["HOME"]
+        else:
+            home_dir = pwd.getpwnam(getpass.getuser()).pw_dir
+        self.assertEqual(home_dir + os.sep + '.bind10' + os.sep,
+                         bindcmd.BindCmdInterpreter().csv_file_dir)
+
+        new_csv_dir = '/something/different/'
+        custom_cmd = bindcmd.BindCmdInterpreter(csv_file_dir=new_csv_dir)
+        self.assertEqual(new_csv_dir, custom_cmd.csv_file_dir)
+
     def test_get_saved_user_info(self):
-        cmd = FakeBindCmdInterpreter()
-        users = cmd._get_saved_user_info('/notexist', 'cvs_file.cvs')
+        cmd = bindcmd.BindCmdInterpreter()
+        users = cmd._get_saved_user_info('/notexist', 'csv_file.csv')
         self.assertEqual([], users)
         
         csvfilename = 'csv_file.csv'
@@ -360,6 +370,40 @@ class TestBindCmdInterpreter(unittest.TestCase):
         self.assertEqual([], users)
         os.remove(csvfilename)
 
+
+class TestCommandLineOptions(unittest.TestCase):
+    class FakeParserError(Exception):
+        """An exception thrown from FakeOptionParser on parser error.
+        """
+        pass
+
+    class FakeOptionParser(OptionParser):
+        """This fake class emulates the OptionParser class with customized
+        error handling for the convenient of tests.
+        """
+        def __init__(self):
+            OptionParser.__init__(self)
+
+        def error(self, msg):
+            raise TestCommandLineOptions.FakeParserError
+
+    def setUp(self):
+        self.parser = self.FakeOptionParser()
+        set_bindctl_options(self.parser)
+
+    def test_csv_file_dir(self):
+        # by default the option is "undefined"
+        (options, _) = self.parser.parse_args([])
+        self.assertEqual(None, options.csv_file_dir)
+
+        # specify the option, valid case.
+        (options, _) = self.parser.parse_args(['--csv-file-dir', 'some_dir'])
+        self.assertEqual('some_dir', options.csv_file_dir)
+
+        # missing option arg; should trigger parser error.
+        self.assertRaises(self.FakeParserError, self.parser.parse_args,
+                          ['--csv-file-dir'])
+
 if __name__== "__main__":
     unittest.main()
     

src/bin/resolver/Makefile.am

@@ -51,6 +51,7 @@ b10_resolver_LDADD += $(top_builddir)/src/lib/log/liblog.la
 b10_resolver_LDADD += $(top_builddir)/src/lib/server_common/libserver_common.la
 b10_resolver_LDADD += $(top_builddir)/src/lib/cache/libcache.la
 b10_resolver_LDADD += $(top_builddir)/src/lib/nsas/libnsas.la
+b10_resolver_LDADD += $(top_builddir)/src/lib/resolve/libresolve.la
 b10_resolver_LDADD += $(top_builddir)/src/bin/auth/change_user.o
 b10_resolver_LDFLAGS = -pthread
 

src/bin/resolver/main.cc

@@ -45,6 +45,9 @@
 #include <resolver/spec_config.h>
 #include <resolver/resolver.h>
 
+#include <cache/resolver_cache.h>
+#include <nsas/nameserver_address_store.h>
+
 #include <log/dummylog.h>
 
 using namespace std;
@@ -59,7 +62,7 @@ namespace {
 static const string PROGRAM = "Resolver";
 
 IOService io_service;
-static Resolver *resolver;
+static boost::shared_ptr<Resolver> resolver;
 
 ConstElementPtr
 my_config_handler(ConstElementPtr new_config) {
@@ -135,15 +138,58 @@ main(int argc, char* argv[]) {
             specfile = string(RESOLVER_SPECFILE_LOCATION);
         }
 
-        resolver = new Resolver();
+        resolver = boost::shared_ptr<Resolver>(new Resolver());
         dlog("Server created.");
 
         SimpleCallback* checkin = resolver->getCheckinProvider();
         DNSLookup* lookup = resolver->getDNSLookupProvider();
         DNSAnswer* answer = resolver->getDNSAnswerProvider();
 
+        isc::nsas::NameserverAddressStore nsas(resolver);
+        resolver->setNameserverAddressStore(nsas);
+
+        isc::cache::ResolverCache cache;
+        resolver->setCache(cache);
+        
+        // TODO priming query, remove root from direct
+        // Fake a priming query result here (TODO2 how to flag non-expiry?)
+        // propagation to runningquery. And check for forwarder mode?
+        isc::dns::QuestionPtr root_question(new isc::dns::Question(
+                                            isc::dns::Name("."),
+                                            isc::dns::RRClass::IN(),
+                                            isc::dns::RRType::NS()));
+        isc::dns::RRsetPtr root_ns_rrset(new isc::dns::RRset(isc::dns::Name("."), 
+                                         isc::dns::RRClass::IN(),
+                                         isc::dns::RRType::NS(),
+                                         isc::dns::RRTTL(8888)));
+        root_ns_rrset->addRdata(isc::dns::rdata::createRdata(isc::dns::RRType::NS(),
+                                                             isc::dns::RRClass::IN(),
+                                                             "l.root-servers.net."));
+        isc::dns::RRsetPtr root_a_rrset(new isc::dns::RRset(isc::dns::Name("l.root-servers.net"), 
+                                        isc::dns::RRClass::IN(),
+                                        isc::dns::RRType::A(),
+                                        isc::dns::RRTTL(8888)));
+        root_a_rrset->addRdata(isc::dns::rdata::createRdata(isc::dns::RRType::A(),
+                                                             isc::dns::RRClass::IN(),
+                                                             "199.7.83.42"));
+        isc::dns::RRsetPtr root_aaaa_rrset(new isc::dns::RRset(isc::dns::Name("l.root-servers.net"), 
+                                        isc::dns::RRClass::IN(),
+                                        isc::dns::RRType::AAAA(),
+                                        isc::dns::RRTTL(8888)));
+        root_aaaa_rrset->addRdata(isc::dns::rdata::createRdata(isc::dns::RRType::AAAA(),
+                                                             isc::dns::RRClass::IN(),
+                                                             "2001:500:3::42"));
+        isc::dns::MessagePtr priming_result(new isc::dns::Message(isc::dns::Message::RENDER));
+        priming_result->addQuestion(root_question);
+        priming_result->addRRset(isc::dns::Message::SECTION_ANSWER, root_ns_rrset);
+        priming_result->addRRset(isc::dns::Message::SECTION_ADDITIONAL, root_a_rrset);
+        priming_result->addRRset(isc::dns::Message::SECTION_ADDITIONAL, root_aaaa_rrset);
+        cache.update(*priming_result);
+        cache.update(root_ns_rrset);
+        cache.update(root_a_rrset);
+        cache.update(root_aaaa_rrset);
+        
         DNSService dns_service(io_service, checkin, lookup, answer);
-
         resolver->setDNSService(dns_service);
         dlog("IOService created.");
 
@@ -172,7 +218,6 @@ main(int argc, char* argv[]) {
 
     delete config_session;
     delete cc_session;
-    delete resolver;
 
     return (ret);
 }

src/bin/resolver/resolver.cc

@@ -41,6 +41,8 @@
 #include <dns/messagerenderer.h>
 #include <server_common/portconfig.h>
 
+#include <resolve/recursive_query.h>
+
 #include <log/dummylog.h>
 
 #include <resolver/resolver.h>
@@ -74,10 +76,15 @@ public:
         queryShutdown();
     }
 
-    void querySetup(DNSService& dnss) {
+    void querySetup(DNSService& dnss,
+                    isc::nsas::NameserverAddressStore& nsas,
+                    isc::cache::ResolverCache& cache)
+    {
         assert(!rec_query_); // queryShutdown must be called first
         dlog("Query setup");
-        rec_query_ = new RecursiveQuery(dnss, upstream_,
+        rec_query_ = new RecursiveQuery(dnss, 
+                                        nsas, cache,
+                                        upstream_,
                                         upstream_root_,
                                         query_timeout_,
                                         client_timeout_,
@@ -129,7 +136,7 @@ public:
             }
         }
     }
-
+    
     void resolve(const isc::dns::QuestionPtr& question,
         const isc::resolve::ResolverInterface::CallbackPtr& callback);
 
@@ -342,6 +349,19 @@ Resolver::setDNSService(asiolink::DNSService& dnss) {
 }
 
 void
+Resolver::setNameserverAddressStore(isc::nsas::NameserverAddressStore& nsas)
+{
+    nsas_ = &nsas;
+}
+
+void
+Resolver::setCache(isc::cache::ResolverCache& cache)
+{
+    cache_ = &cache;
+}
+
+
+void
 Resolver::setConfigSession(ModuleCCSession* config_session) {
     impl_->config_session_ = config_session;
 }
@@ -544,7 +564,7 @@ Resolver::updateConfig(ConstElementPtr config) {
 
         if (need_query_restart) {
             impl_->queryShutdown();
-            impl_->querySetup(*dnss_);
+            impl_->querySetup(*dnss_, *nsas_, *cache_);
         }
         return (isc::config::createAnswer());
     } catch (const isc::Exception& error) {

src/bin/resolver/resolver.h

@@ -24,6 +24,9 @@
 
 #include <asiolink/asiolink.h>
 
+#include <nsas/nameserver_address_store.h>
+#include <cache/resolver_cache.h>
+
 #include <resolve/resolver_interface.h>
 
 class ResolverImpl;
@@ -86,10 +89,26 @@ public:
 
     /// \brief Assign an ASIO IO Service queue to this Resolver object
     void setDNSService(asiolink::DNSService& dnss);
+    
+    /// \brief Assign a NameserverAddressStore to this Resolver object
+    void setNameserverAddressStore(isc::nsas::NameserverAddressStore &nsas);
+    
+    /// \brief Assign a cache to this Resolver object
+    void setCache(isc::cache::ResolverCache& cache);
 
     /// \brief Return this object's ASIO IO Service queue
     asiolink::DNSService& getDNSService() const { return (*dnss_); }
 
+    /// \brief Returns this object's NSAS
+    isc::nsas::NameserverAddressStore& getNameserverAddressStore() const {
+        return *nsas_;
+    };
+
+    /// \brief Returns this object's ResolverCache
+    isc::cache::ResolverCache& getResolverCache() const {
+        return *cache_;
+    };
+    
     /// \brief Return pointer to the DNS Lookup callback function
     asiolink::DNSLookup* getDNSLookupProvider() { return (dns_lookup_); }
 
@@ -208,6 +227,8 @@ private:
     asiolink::SimpleCallback* checkin_;
     asiolink::DNSLookup* dns_lookup_;
     asiolink::DNSAnswer* dns_answer_;
+    isc::nsas::NameserverAddressStore* nsas_;
+    isc::cache::ResolverCache* cache_;
 };
 
 #endif // __RESOLVER_H

src/bin/resolver/tests/Makefile.am

@@ -40,6 +40,7 @@ run_unittests_LDADD += $(top_builddir)/src/lib/log/liblog.la
 run_unittests_LDADD += $(top_builddir)/src/lib/server_common/libserver_common.la
 run_unittests_LDADD += $(top_builddir)/src/lib/cache/libcache.la
 run_unittests_LDADD += $(top_builddir)/src/lib/nsas/libnsas.la
+run_unittests_LDADD += $(top_builddir)/src/lib/resolve/libresolve.la
 
 # Note the ordering matters: -Wno-... must follow -Wextra (defined in
 # B10_CXXFLAGS

src/lib/Makefile.am

@@ -1,2 +1,2 @@
-SUBDIRS = exceptions dns cc config datasrc python xfr bench log \
-          resolve nsas cache asiolink testutils server_common
+SUBDIRS = exceptions dns cc config python xfr bench log asiolink \
+          nsas cache resolve testutils datasrc server_common

src/lib/asiolink/Makefile.am

@@ -13,41 +13,39 @@ CLEANFILES = *.gcno *.gcda
 # which would make the build fail with -Werror (our default setting).
 lib_LTLIBRARIES = libasiolink.la
 libasiolink_la_SOURCES  = asiolink.h
+libasiolink_la_SOURCES += asiolink_utilities.h
+libasiolink_la_SOURCES += asiodef.cc asiodef.h
 libasiolink_la_SOURCES += dns_answer.h
 libasiolink_la_SOURCES += dns_lookup.h
 libasiolink_la_SOURCES += dns_server.h
-libasiolink_la_SOURCES += dns_service.h dns_service.cc
+libasiolink_la_SOURCES += dns_service.cc dns_service.h
 libasiolink_la_SOURCES += dummy_io_cb.h
-libasiolink_la_SOURCES += interval_timer.h interval_timer.cc
-libasiolink_la_SOURCES += io_address.h io_address.cc
+libasiolink_la_SOURCES += interval_timer.cc interval_timer.h
+libasiolink_la_SOURCES += io_address.cc io_address.h
 libasiolink_la_SOURCES += io_asio_socket.h
-libasiolink_la_SOURCES += io_endpoint.h io_endpoint.cc
+libasiolink_la_SOURCES += io_endpoint.cc io_endpoint.h
 libasiolink_la_SOURCES += io_error.h
-libasiolink_la_SOURCES += io_fetch.h io_fetch.cc
+libasiolink_la_SOURCES += io_fetch.cc io_fetch.h
 libasiolink_la_SOURCES += io_message.h
+libasiolink_la_SOURCES += qid_gen.cc qid_gen.h
 libasiolink_la_SOURCES += io_service.h io_service.cc
 libasiolink_la_SOURCES += io_socket.h io_socket.cc
-libasiolink_la_SOURCES += recursive_query.h recursive_query.cc
 libasiolink_la_SOURCES += simple_callback.h
 libasiolink_la_SOURCES += tcp_endpoint.h
-libasiolink_la_SOURCES += tcp_server.h tcp_server.cc
+libasiolink_la_SOURCES += tcp_server.cc tcp_server.h
 libasiolink_la_SOURCES += tcp_socket.h
 libasiolink_la_SOURCES += udp_endpoint.h
-libasiolink_la_SOURCES += udp_server.h udp_server.cc
+libasiolink_la_SOURCES += udp_server.cc udp_server.h
 libasiolink_la_SOURCES += udp_socket.h
-libasiolink_la_SOURCES += qid_gen.cc qid_gen.h
+
+EXTRA_DIST = asiodef.msg
+
 # Note: the ordering matters: -Wno-... must follow -Wextra (defined in
 # B10_CXXFLAGS)
 libasiolink_la_CXXFLAGS = $(AM_CXXFLAGS)
-if USE_GXX
-libasiolink_la_CXXFLAGS += -Wno-unused-parameter
-endif
 if USE_CLANGPP
 # Same for clang++, but we need to turn off -Werror completely.
 libasiolink_la_CXXFLAGS += -Wno-error
 endif
 libasiolink_la_CPPFLAGS = $(AM_CPPFLAGS)
 libasiolink_la_LIBADD = $(top_builddir)/src/lib/log/liblog.la
-libasiolink_la_LIBADD += $(top_builddir)/src/lib/resolve/libresolve.la
-libasiolink_la_LIBADD += $(top_builddir)/src/lib/cache/libcache.la
-libasiolink_la_LIBADD += $(top_builddir)/src/lib/nsas/libnsas.la

src/lib/asiolink/asiodef.cc

@@ -0,0 +1,37 @@
+// File created from asiodef.msg on Mon Feb 28 17:15:30 2011
+
+#include <cstddef>
+#include <log/message_types.h>
+#include <log/message_initializer.h>
+
+namespace asiolink {
+
+extern const isc::log::MessageID ASIO_FETCHCOMP = "FETCHCOMP";
+extern const isc::log::MessageID ASIO_FETCHSTOP = "FETCHSTOP";
+extern const isc::log::MessageID ASIO_OPENSOCK = "OPENSOCK";
+extern const isc::log::MessageID ASIO_RECVSOCK = "RECVSOCK";
+extern const isc::log::MessageID ASIO_RECVTMO = "RECVTMO";
+extern const isc::log::MessageID ASIO_SENDSOCK = "SENDSOCK";
+extern const isc::log::MessageID ASIO_UNKORIGIN = "UNKORIGIN";
+extern const isc::log::MessageID ASIO_UNKRESULT = "UNKRESULT";
+
+} // namespace asiolink
+
+namespace {
+
+const char* values[] = {
+    "FETCHCOMP", "upstream fetch to %s(%d) has now completed",
+    "FETCHSTOP", "upstream fetch to %s(%d) has been stopped",
+    "OPENSOCK", "error %d opening %s socket to %s(%d)",
+    "RECVSOCK", "error %d reading %s data from %s(%d)",
+    "RECVTMO", "receive timeout while waiting for data from %s(%d)",
+    "SENDSOCK", "error %d sending data using %s to %s(%d)",
+    "UNKORIGIN", "unknown origin for ASIO error code %d (protocol: %s, address %s)",
+    "UNKRESULT", "unknown result (%d) when IOFetch::stop() was executed for I/O to %s(%d)",
+    NULL
+};
+
+const isc::log::MessageInitializer initializer(values);
+
+} // Anonymous namespace
+

src/lib/asiolink/asiodef.h

@@ -0,0 +1,21 @@
+// File created from asiodef.msg on Mon Feb 28 17:15:30 2011
+
+#ifndef __ASIODEF_H
+#define __ASIODEF_H
+
+#include <log/message_types.h>
+
+namespace asiolink {
+
+extern const isc::log::MessageID ASIO_FETCHCOMP;
+extern const isc::log::MessageID ASIO_FETCHSTOP;
+extern const isc::log::MessageID ASIO_OPENSOCK;
+extern const isc::log::MessageID ASIO_RECVSOCK;
+extern const isc::log::MessageID ASIO_RECVTMO;
+extern const isc::log::MessageID ASIO_SENDSOCK;
+extern const isc::log::MessageID ASIO_UNKORIGIN;
+extern const isc::log::MessageID ASIO_UNKRESULT;
+
+} // namespace asiolink
+
+#endif // __ASIODEF_H

src/lib/asiolink/asiodef.msg

@@ -0,0 +1,56 @@
+# Copyright (C) 2011  Internet Systems Consortium, Inc. ("ISC")
+#
+# Permission to use, copy, modify, and/or distribute this software for any
+# purpose with or without fee is hereby granted, provided that the above
+# copyright notice and this permission notice appear in all copies.
+#
+# THE SOFTWARE IS PROVIDED "AS IS" AND ISC DISCLAIMS ALL WARRANTIES WITH
+# REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
+# AND FITNESS.  IN NO EVENT SHALL ISC BE LIABLE FOR ANY SPECIAL, DIRECT,
+# INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
+# LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
+# OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
+# PERFORMANCE OF THIS SOFTWARE.
+
+$PREFIX ASIO_
+$NAMESPACE asiolink
+
+FETCHCOMP   upstream fetch to %s(%d) has now completed
++ A debug message, this records the the upstream fetch (a query made by the
++ resolver on behalf of its client) to the specified address has completed.
+
+FETCHSTOP   upstream fetch to %s(%d) has been stopped
++ An external component has requested the halting of an upstream fetch.  This
++ is an allowed operation, and the message should only appear if debug is
++ enabled.
+
+OPENSOCK    error %d opening %s socket to %s(%d)
++ The asynchronous I/O code encountered an error when trying to open a socket
++ of the specified protocol in order to send a message to the target address.
++ The the number of the system error that cause the problem is given in the
++ message.
+
+RECVSOCK    error %d reading %s data from %s(%d)
++ The asynchronous I/O code encountered an error when trying read data from
++ the specified address on the given protocol.  The the number of the system
++ error that cause the problem is given in the message.
+
+SENDSOCK    error %d sending data using %s to %s(%d)
++ The asynchronous I/O code encountered an error when trying send data to
++ the specified address on the given protocol.  The the number of the system
++ error that cause the problem is given in the message.
+
+RECVTMO     receive timeout while waiting for data from %s(%d)
++ An upstream fetch from the specified address timed out.  This may happen for
++ any number of reasons and is most probably a problem at the remote server
++ or a problem on the network.  The message will only appear if debug is
++ enabled.
+
+UNKORIGIN  unknown origin for ASIO error code %d (protocol: %s, address %s)
++ This message should not appear and indicates an internal error if it does.
++ Please enter a bug report.
+
+UNKRESULT  unknown result (%d) when IOFetch::stop() was executed for I/O to %s(%d)
++ The termination method of the resolver's upstream fetch class was called with
++ an unknown result code (which is given in the message).  This message should
++ not appear and may indicate an internal error.  Please enter a bug report.

src/lib/asiolink/asiolink.h

@@ -25,7 +25,6 @@
 #include <asiolink/dns_lookup.h>
 #include <asiolink/dns_answer.h>
 #include <asiolink/simple_callback.h>
-#include <asiolink/recursive_query.h>
 #include <asiolink/interval_timer.h>
 
 #include <asiolink/io_address.h>
@@ -85,7 +84,3 @@
 /// http://think-async.com/Asio/asio-1.3.1/doc/asio/reference/asio_handler_allocate.html
 
 #endif // __ASIOLINK_H
-
-// Local Variables: 
-// mode: c++
-// End: 

src/lib/asiolink/asiolink_utilities.h

@@ -0,0 +1,61 @@
+// Copyright (C) 2011  Internet Systems Consortium, Inc. ("ISC")
+//
+// Permission to use, copy, modify, and/or distribute this software for any
+// purpose with or without fee is hereby granted, provided that the above
+// copyright notice and this permission notice appear in all copies.
+//
+// THE SOFTWARE IS PROVIDED "AS IS" AND ISC DISCLAIMS ALL WARRANTIES WITH
+// REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
+// AND FITNESS.  IN NO EVENT SHALL ISC BE LIABLE FOR ANY SPECIAL, DIRECT,
+// INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
+// LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
+// OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
+// PERFORMANCE OF THIS SOFTWARE.
+
+#ifndef __ASIOLINK_UTILITIES_H
+#define __ASIOLINK_UTILITIES_H
+
+#include <cstddef>
+
+namespace asiolink {
+
+/// \brief Read Unsigned 16-Bit Integer from Buffer
+///
+/// This is essentially a copy of the isc::dns::InputBuffer::readUint16.  It
+/// should really be moved into a separate library.
+///
+/// \param buffer Data buffer at least two bytes long of which the first two
+///        bytes are assumed to represent a 16-bit integer in network-byte
+///        order.
+///
+/// \return Value of 16-bit integer
+inline uint16_t
+readUint16(const void* buffer) {
+    const uint8_t* byte_buffer = static_cast<const uint8_t*>(buffer);
+
+    uint16_t result = (static_cast<uint16_t>(byte_buffer[0])) << 8;
+    result |= static_cast<uint16_t>(byte_buffer[1]);
+
+    return (result);
+}
+
+/// \brief Write Unisgned 16-Bit Integer to Buffer
+///
+/// This is essentially a copy of isc::dns::OutputBuffer::writeUint16.  It
+/// should really be moved into a separate library.
+///
+/// \param value 16-bit value to convert
+/// \param buffer Data buffer at least two bytes long into which the 16-bit
+///        value is written in network-byte order.
+
+inline void
+writeUint16(uint16_t value, void* buffer) {
+    uint8_t* byte_buffer = static_cast<uint8_t*>(buffer);
+
+    byte_buffer[0] = static_cast<uint8_t>((value & 0xff00U) >> 8);
+    byte_buffer[1] = static_cast<uint8_t>(value & 0x00ffU);
+}
+
+} // namespace asiolink
+
+#endif // __ASIOLINK_UTILITIES_H

src/lib/asiolink/dns_server.h

@@ -21,7 +21,7 @@ namespace asiolink {
 
 /// \brief The \c DNSServer class is a wrapper (and base class) for
 /// classes which provide DNS server functionality.
-/// 
+///
 /// The classes derived from this one, \c TCPServer and \c UDPServer,
 /// act as the interface layer between clients sending queries, and
 /// functions defined elsewhere that provide answers to those queries.
@@ -42,10 +42,10 @@ namespace asiolink {
 /// when "forking", and that instances will be posted as ASIO handler
 /// objects, which are always copied.
 ///
-/// Because these objects are frequently copied, it is recommended 
+/// Because these objects are frequently copied, it is recommended
 /// that derived classes be kept small to reduce copy overhead.
 class DNSServer {
-protected: 
+protected:
     ///
     /// \name Constructors and destructors
     ///
@@ -66,7 +66,7 @@ public:
     /// the ones in the derived class.  This makes it possible to pass
     /// instances of derived classes as references to this base class
     /// without losing access to derived class data.
-    /// 
+    ///
     //@{
     /// \brief The funtion operator
     virtual void operator()(asio::error_code ec = asio::error_code(),
@@ -87,7 +87,7 @@ public:
 
     /// \brief Indicate whether the server is able to send an answer
     /// to a query.
-    /// 
+    ///
     /// This is presently used only for testing purposes.
     virtual bool hasAnswer() { return (self_->hasAnswer()); }
 

src/lib/asiolink/dns_service.h

@@ -26,13 +26,13 @@ class DNSLookup;
 class DNSAnswer;
 class DNSServiceImpl;
 
+/// \brief Handle DNS Queries
 ///
 /// DNSService is the service that handles DNS queries and answers with
 /// a given IOService. This class is mainly intended to hold all the
 /// logic that is shared between the authoritative and the recursive
 /// server implementations. As such, it handles asio, including config
 /// updates (through the 'Checkinprovider'), and listening sockets.
-/// 
 class DNSService {
     ///
     /// \name Constructors and Destructor

src/lib/asiolink/dummy_io_cb.h

@@ -39,6 +39,14 @@ public:
     /// \brief Asynchronous I/O callback method
     ///
     /// \param error Unused
+    void operator()(asio::error_code)
+    {
+        // TODO: log an error if this method ever gets called.
+    }
+
+    /// \brief Asynchronous I/O callback method
+    ///
+    /// \param error Unused
     /// \param length Unused
     void operator()(asio::error_code, size_t)
     {

src/lib/asiolink/internal/iofetch.h

@@ -1,125 +0,0 @@
-// Copyright (C) 2010  Internet Systems Consortium, Inc. ("ISC")
-//
-// Permission to use, copy, modify, and/or distribute this software for any
-// purpose with or without fee is hereby granted, provided that the above
-// copyright notice and this permission notice appear in all copies.
-//
-// THE SOFTWARE IS PROVIDED "AS IS" AND ISC DISCLAIMS ALL WARRANTIES WITH
-// REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
-// AND FITNESS.  IN NO EVENT SHALL ISC BE LIABLE FOR ANY SPECIAL, DIRECT,
-// INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
-// LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
-// OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
-// PERFORMANCE OF THIS SOFTWARE.
-
-#ifndef __IOFETCH_H
-#define __IOFETCH_H 1
-
-#include <config.h>
-
-#include <asio.hpp>
-#include <boost/shared_array.hpp>
-#include <boost/shared_ptr.hpp>
-
-#include <dns/buffer.h>
-#include <dns/message.h>
-#include <dns/messagerenderer.h>
-
-#include <asiolink/asiolink.h>
-#include <asiolink/internal/coroutine.h>
-
-// This file contains TCP/UDP-specific implementations of generic classes 
-// defined in asiolink.h.  It is *not* intended to be part of the public
-// API.
-
-namespace asiolink {
-//
-// Asynchronous UDP/TCP coroutine for upstream fetches
-//
-//class IOFetch : public coroutine, public UdpFetch, public TcpFetch {
-class IOFetch : public coroutine {
-public:
-    // TODO Maybe this should be more generic than just for IOFetch?
-    ///
-    /// \brief Result of the query
-    ///
-    /// This is related only to contacting the remote server. If the answer
-    ///indicates error, it is still counted as SUCCESS here, if it comes back.
-    ///
-    enum Result {
-        SUCCESS,
-        TIME_OUT,
-        STOPPED
-    };
-    /// Abstract callback for the IOFetch.
-    class Callback {
-    public:
-        virtual ~Callback() {}
-
-        /// This will be called when the IOFetch is completed
-        virtual void operator()(Result result) = 0;
-    };
-    ///
-    /// \brief Constructor.
-    ///
-    /// It creates the query.
-    /// @param callback will be called when we terminate. It is your task to
-    ///        delete it if allocated on heap.
-    ///@param timeout in ms.
-    ///
-    IOFetch(asio::io_service& io_service,
-                      const isc::dns::Question& q,
-                      const IOAddress& addr, uint16_t port,
-                      isc::dns::OutputBufferPtr buffer,
-                      Callback* callback, int timeout = -1, 
-                      int protocol = IPPROTO_UDP);
-    void operator()(asio::error_code ec = asio::error_code(),
-                    size_t length = 0);
-    /// Terminate the query.
-    void stop(Result reason = STOPPED);
-private:
-    enum { MAX_LENGTH = 4096 };
-
-    ///
-    /// \short Private data
-    ///
-    /// They are not private because of stability of the
-    /// interface (this is private class anyway), but because this class
-    /// will be copyed often (it is used as a coroutine and passed as callback
-    /// to many async_*() functions) and we want keep the same data. Some of
-    /// the data is not copyable too.
-    ///
-    //struct IOFetchProtocol;
-    //boost::shared_ptr<IOFetchProtocol> data_;
-    //struct UdpData;
-    //struct TcpData;
-    boost::shared_ptr<UdpFetch> data_;
-    boost::shared_ptr<TcpFetch> tcp_data_;
-};
-class UdpFetch : public IOFetch {
-    public:
-        struct UdpData;
-        explicit UdpFetch(asio::io_service& io_service,
-                          const isc::dns::Question& q,
-                          const IOAddress& addr,
-                          uint16_t port,
-                          isc::dns::OutputBufferPtr buffer,
-                          IOFetch::Callback *callback,
-                          int timeout);
-};
-class TcpFetch : public IOFetch {
-    public:
-        struct TcpData;
-        explicit TcpFetch(io_service& io_service, const Question& q,
-                 const IOAddress& addr, uint16_t port,
-                 OutputBufferPtr buffer, Callback *callback, int timeout);
-};
-
-}
-
-
-#endif // __IOFETCH_H
-
-// Local Variables: 
-// mode: c++
-// End: 

src/lib/asiolink/interval_timer.h

@@ -37,7 +37,7 @@ struct IntervalTimerImpl;
 /// The function calls the call back function set by \c setup() and updates
 /// the timer to expire in (now + interval) milliseconds.
 /// The type of call back function is \c void(void).
-/// 
+///
 /// The call back function will not be called if the instance of this class is
 /// destroyed before the timer is expired.
 ///

src/lib/asiolink/io_address.h

@@ -121,7 +121,3 @@ private:
 
 }      // asiolink
 #endif // __IO_ADDRESS_H
-
-// Local Variables: 
-// mode: c++
-// End: 

src/lib/asiolink/io_asio_socket.h

@@ -26,6 +26,8 @@
 #include <exceptions/exceptions.h>
 #include <coroutine.h>
 
+#include <dns/buffer.h>
+
 #include <asiolink/io_error.h>
 #include <asiolink/io_socket.h>
 
@@ -41,7 +43,24 @@ public:
         IOError(file, line, what) {}
 };
 
+/// \brief Error setting socket options
+///
+/// Thrown if attempt to change socket options fails.
+class SocketSetError : public IOError {
+public:
+    SocketSetError(const char* file, size_t line, const char* what) :
+        IOError(file, line, what) {}
+};
 
+/// \brief Buffer overflow
+///
+/// Thrown if an attempt is made to receive into an area beyond the end of
+/// the receive data buffer.
+class BufferOverflow : public IOError {
+public:
+    BufferOverflow(const char* file, size_t line, const char* what) :
+        IOError(file, line, what) {}
+};
 
 /// Forward declaration of an IOEndpoint
 class IOEndpoint;
@@ -91,24 +110,23 @@ public:
 
     /// \brief Return the "native" representation of the socket.
     ///
-    /// In practice, this is the file descriptor of the socket for
-    /// UNIX-like systems so the current implementation simply uses
-    /// \c int as the type of the return value.
-    /// We may have to need revisit this decision later.
+    /// In practice, this is the file descriptor of the socket for UNIX-like
+    /// systems so the current implementation simply uses \c int as the type of
+    /// the return value. We may have to need revisit this decision later.
     ///
-    /// In general, the application should avoid using this method;
-    /// it essentially discloses an implementation specific "handle" that
-    /// can change the internal state of the socket (consider the
-    /// application closes it, for example).
-    /// But we sometimes need to perform very low-level operations that
-    /// requires the native representation.  Passing the file descriptor
-    /// to a different process is one example.
-    /// This method is provided as a necessary evil for such limited purposes.
+    /// In general, the application should avoid using this method; it
+    /// essentially discloses an implementation specific "handle" that can
+    /// change the internal state of the socket (consider what would happen if
+    /// the application closes it, for example).  But we sometimes need to
+    /// perform very low-level operations that requires the native
+    /// representation.  Passing the file descriptor to a different process is
+    /// one example.  This method is provided as a necessary evil for such
+    /// limited purposes.
     ///
     /// This method never throws an exception.
     ///
     /// \return The native representation of the socket.  This is the socket
-    /// file descriptor for UNIX-like systems.
+    ///         file descriptor for UNIX-like systems.
     virtual int getNative() const = 0;
 
     /// \brief Return the transport protocol of the socket.
@@ -118,36 +136,50 @@ public:
     ///
     /// This method never throws an exception.
     ///
-    /// \return IPPROTO_UDP for UDP sockets
-    /// \return IPPROTO_TCP for TCP sockets
+    /// \return \c IPPROTO_UDP for UDP sockets, \c IPPROTO_TCP for TCP sockets
     virtual int getProtocol() const = 0;
 
-    /// \brief Open AsioSocket
+    /// \brief Is Open() synchronous?
     ///
-    /// Opens the socket for asynchronous I/O.  On a UDP socket, this is merely
-    /// an "open()" on the underlying socket (so completes immediately), but on
-    /// a TCP socket it also connects to the remote end (which is done as an
-    /// asynchronous operation).
+    /// On a TCP socket, an "open" operation is a call to the socket's "open()"
+    /// method followed by a connection to the remote system: it is an
+    /// asynchronous operation.  On a UDP socket, it is just a call to "open()"
+    /// and completes synchronously.
     ///
     /// For TCP, signalling of the completion of the operation is done by
     /// by calling the callback function in the normal way.  This could be done
     /// for UDP (by posting en event on the event queue); however, that will
-    /// incur additional overhead in the most common case.  Instead, the return
-    /// value indicates whether the operation was asynchronous or not. If yes,
-    /// (i.e. TCP) the callback has been posted to the event queue: if no (UDP),
-    /// no callback has been posted (in which case it is up to the caller as to
-    /// whether they want to manually post the callback themself.)
+    /// incur additional overhead in the most common case.  So we give the
+    /// caller the choice for calling this open() method synchronously or
+    /// asynchronously.
+    ///
+    /// Owing to the way that the stackless coroutines are implemented, we need
+    /// to know _before_ executing the "open" function whether or not it is
+    /// asynchronous.  So this method is called to provide that information.
+    ///
+    /// (The reason there is a need to know is because the call to open() passes
+    /// in the state of the coroutine at the time the call is made.  On an
+    /// asynchronous I/O, we need to set the state to point to the statement
+    /// after the call to open() _before_ we pass the corouine to the open()
+    /// call.  Unfortunately, the macros that set the state of the coroutine
+    /// also yield control - which we don't want to do if the open is
+    /// synchronous.  Hence we need to know before we make the call to open()
+    /// whether that call will complete asynchronously.)
+    virtual bool isOpenSynchronous() const = 0;
+
+    /// \brief Open AsioSocket
+    ///
+    /// Opens the socket for asynchronous I/O.  The open will complete
+    /// synchronously on UCP or asynchronously on TCP (in which case a callback
+    /// will be queued).
     ///
     /// \param endpoint Pointer to the endpoint object.  This is ignored for
-    /// a UDP socket (the target is specified in the send call), but should
-    /// be of type TCPEndpoint for a TCP connection.
+    ///        a UDP socket (the target is specified in the send call), but
+    ///        should be of type TCPEndpoint for a TCP connection.
     /// \param callback I/O Completion callback, called when the operation has
-    /// completed, but only if the operation was asynchronous.
-    ///
-    /// \return true if an asynchronous operation was started and the caller
-    /// should yield and wait for completion, false if the operation was
-    /// completed synchronously and no callback was queued.
-    virtual bool open(const IOEndpoint* endpoint, C& callback) = 0;
+    ///        completed, but only if the operation was asynchronous. (It is
+    ///        ignored on a UDP socket.)
+    virtual void open(const IOEndpoint* endpoint, C& callback) = 0;
 
     /// \brief Send Asynchronously
     ///
@@ -160,44 +192,85 @@ public:
     /// \param endpoint Target of the send
     /// \param callback Callback object.
     virtual void asyncSend(const void* data, size_t length,
-        const IOEndpoint* endpoint, C& callback) = 0;
+                           const IOEndpoint* endpoint, C& callback) = 0;
 
     /// \brief Receive Asynchronously
     ///
-    /// This correstponds to async_receive_from() for UDP sockets and
+    /// This corresponds to async_receive_from() for UDP sockets and
     /// async_receive() for TCP.  In both cases, an endpoint argument is
     /// supplied to receive the source of the communication.  For TCP it will
     /// be filled in with details of the connection.
     ///
     /// \param data Buffer to receive incoming message
     /// \param length Length of the data buffer
-    /// \param cumulative Amount of data that should already be in the buffer.
+    /// \param offset Offset into buffer where data is to be put.  Although the
+    ///        offset could be implied by adjusting "data" and "length"
+    ///        appropriately, using this argument allows data to be specified as
+    ///        "const void*" - the overhead of converting it to a pointer to a
+    ///        set of bytes is hidden away here.
     /// \param endpoint Source of the communication
     /// \param callback Callback object
-    virtual void asyncReceive(void* data, size_t length, size_t cumulative,
-        IOEndpoint* endpoint, C& callback) = 0;
-
-    /// \brief Checks if the data received is complete.
-    ///
-    /// This applies to TCP receives, where the data is a byte stream and a
-    /// receive is not guaranteed to receive the entire message.  DNS messages
-    /// over TCP are prefixed by a two-byte count field.  This method takes the
-    /// amount received so far and the amount received in this I/O and checks
-    /// if the message is complete, returning the appropriate indication.  As
-    /// a side-effect, it also updates the amount received.
-    ///
-    /// For a UDP receive, all the data is received in one I/O, so this is
-    /// effectively a no-op (although it does update the amount received).
-    ///
-    /// \param data Data buffer containing data to date
-    /// \param length Amount of data received in last asynchronous I/O
-    /// \param cumulative On input, amount of data received before the last
-    /// I/O.  On output, the total amount of data received to date.
+    virtual void asyncReceive(void* data, size_t length, size_t offset,
+                              IOEndpoint* endpoint, C& callback) = 0;
+
+    /// \brief Processes received data
+    ///
+    /// In the IOFetch code, data is received into a staging buffer before being
+    /// copied into the target buffer.  (This is because (a) we don't know how
+    /// much data we will be receiving, so don't know how to size the output
+    /// buffer and (b) TCP data is preceded by a two-byte count field that needs
+    /// to be discarded before being returned to the user.)
+    ///
+    /// An additional consideration is that TCP data is not received in one
+    /// I/O - it may take a number of I/Os - each receiving any non-zero number
+    /// of bytes - to read the entire message.
+    ///
+    /// So the IOFetch code has to loop until it determines that all the data
+    /// has been read.  This is where this method comes in.  It has several
+    /// functions:
+    ///
+    /// - It checks if the received data is complete.
+    /// - If data is not complete, decides if the next set of data is to go into
+    ///   the start of the staging buffer or at some offset into it.  (This
+    ///   simplifies the case we could have in a TCP receive where the two-byte
+    ///   count field is received in one-byte chunks: we put off interpreting
+    ///   the count until we have all of it.  The alternative - copying the
+    ///   data to the output buffer and interpreting the count from there -
+    ///   would require moving the data in the output buffer by two bytes before
+    ///   returning it to the caller.)
+    /// - Copies data from the staging buffer into the output buffer.
+    ///
+    /// This functionality mainly applies to TCP receives.  For UDP, all the
+    /// data is received in one I/O, so this just copies the data into the
+    /// output buffer.
+    ///
+    /// \param staging Pointer to the start of the staging buffer.
+    /// \param length Amount of data in the staging buffer.
+    /// \param cumulative Amount of data received before the staging buffer is
+    ///        processed (this includes the TCP count field if appropriate).
+    ///        The value should be set to zero before the receive loop is
+    ///        entered, and it will be updated by this method as required.
+    /// \param offset Offset into the staging buffer where the next read should
+    ///        put the received data.  It should be set to zero before the first
+    ///        call and may be updated by this method.
+    /// \param expected Expected amount of data to be received.  This is
+    ///        really the TCP count field and is set to that value when enough
+    ///        of a TCP message is received.  It should be initialized to -1
+    ///        before the first read is executed.
+    /// \param outbuff Output buffer.  Data in the staging buffer may be copied
+    ///        to this output buffer in the call.
     ///
     /// \return true if the receive is complete, false if another receive is
-    /// needed.
-    virtual bool receiveComplete(void* data, size_t length,
-        size_t& cumulative) = 0;
+    ///         needed.  This is always true for UDP, but for TCP involves
+    ///         checking the amount of data received so far against the amount
+    ///         expected (as indicated by the two-byte count field).  If this
+    ///         method returns false, another read should be queued and data
+    ///         should be read into the staging buffer at offset given by the
+    ///         "offset" parameter.
+    virtual bool processReceivedData(const void* staging, size_t length,
+                                     size_t& cumulative, size_t& offset,
+                                     size_t& expected,
+                                     isc::dns::OutputBufferPtr& outbuff) = 0;
 
     /// \brief Cancel I/O On AsioSocket
     virtual void cancel() = 0;
@@ -244,6 +317,13 @@ public:
     virtual int getProtocol() const { return (protocol_); }
 
 
+    /// \brief Is socket opening synchronous?
+    ///
+    /// \return true - it is for this class.
+    bool isOpenSynchronous() const {
+        return true;
+    }
+
     /// \brief Open AsioSocket
     ///
     /// A call that is a no-op on UDP sockets, this opens a connection to the
@@ -273,21 +353,31 @@ public:
     ///
     /// \param data Unused
     /// \param length Unused
-    /// \param cumulative Unused
+    /// \param offset Unused
     /// \param endpoint Unused
     /// \param callback Unused
-    virtual void asyncReceive(void* data, size_t, size_t, IOEndpoint*, C&) { } 
+    virtual void asyncReceive(void* data, size_t, size_t, IOEndpoint*, C&) {
+    }
+
     /// \brief Checks if the data received is complete.
     ///
-    /// \param data Unused
+    /// \param staging Unused
     /// \param length Unused
     /// \param cumulative Unused
+    /// \param offset Unused.
+    /// \param expected Unused.
+    /// \param outbuff Unused.
     ///
     /// \return Always true
-    virtual bool receiveComplete(void*, size_t, size_t&) {
+    virtual bool receiveComplete(const void* staging, size_t length,
+                                 size_t& cumulative, size_t& offset,
+                                 size_t& expected,
+                                 isc::dns::OutputBufferPtr& outbuff)
+    {
         return (true);
     }
 
+
     /// \brief Cancel I/O On AsioSocket
     ///
     /// Must be supplied as it is abstract in the base class.

src/lib/asiolink/io_endpoint.cc

@@ -22,6 +22,7 @@
 
 #include <asiolink/io_address.h>
 #include <asiolink/io_error.h>
+#include <asiolink/io_endpoint.h>
 #include <asiolink/tcp_endpoint.h>
 #include <asiolink/udp_endpoint.h>
 

src/lib/asiolink/io_endpoint.h

@@ -116,7 +116,3 @@ public:
 
 }      // asiolink
 #endif // __IO_ENDPOINT_H
-
-// Local Variables: 
-// mode: c++
-// End: 

src/lib/asiolink/io_fetch.cc

@@ -19,43 +19,157 @@
 #include <netinet/in.h>
 
 #include <boost/bind.hpp>
+#include <boost/scoped_ptr.hpp>
+#include <boost/date_time/posix_time/posix_time_types.hpp>
 
 #include <dns/message.h>
 #include <dns/messagerenderer.h>
 #include <dns/opcode.h>
 #include <dns/rcode.h>
-#include <log/dummylog.h>
+#include <log/logger.h>
 
 #include <asiolink/qid_gen.h>
 
 #include <asio.hpp>
+#include <asio/deadline_timer.hpp>
+
+#include <asiolink/asiodef.h>
+#include <asiolink/io_address.h>
+#include <asiolink/io_asio_socket.h>
+#include <asiolink/io_endpoint.h>
 #include <asiolink/io_fetch.h>
+#include <asiolink/io_service.h>
+#include <asiolink/tcp_endpoint.h>
+#include <asiolink/tcp_socket.h>
+#include <asiolink/udp_endpoint.h>
+#include <asiolink/udp_socket.h>
 
 using namespace asio;
 using namespace isc::dns;
 using namespace isc::log;
 using namespace std;
 
+namespace asiolink {
 
+/// Use the ASIO logger
+
+isc::log::Logger logger("asiolink");
+
+/// \brief IOFetch Data
+///
+/// The data for IOFetch is held in a separate struct pointed to by a shared_ptr
+/// object.  This is because the IOFetch object will be copied often (it is used
+/// as a coroutine and passed as callback to many async_*() functions) and we
+/// want keep the same data).  Organising the data in this way keeps copying to
+/// a minimum.
+struct IOFetchData {
+
+    // The first two members are shared pointers to a base class because what is
+    // actually instantiated depends on whether the fetch is over UDP or TCP,
+    // which is not known until construction of the IOFetch.  Use of a shared
+    // pointer here is merely to ensure deletion when the data object is deleted.
+    boost::scoped_ptr<IOAsioSocket<IOFetch> > socket;
+                                            ///< Socket to use for I/O
+    boost::scoped_ptr<IOEndpoint> remote;   ///< Where the fetch was sent
+    isc::dns::Question          question;   ///< Question to be asked
+    isc::dns::OutputBufferPtr   msgbuf;     ///< Wire buffer for question
+    isc::dns::OutputBufferPtr   received;   ///< Received data put here
+    IOFetch::Callback*          callback;   ///< Called on I/O Completion
+    asio::deadline_timer        timer;      ///< Timer to measure timeouts
+    IOFetch::Protocol           protocol;   ///< Protocol being used
+    size_t                      cumulative; ///< Cumulative received amount
+    size_t                      expected;   ///< Expected amount of data
+    size_t                      offset;     ///< Offset to receive data
+    bool                        stopped;    ///< Have we stopped running?
+    int                         timeout;    ///< Timeout in ms
+
+    // In case we need to log an error, the origin of the last asynchronous
+    // I/O is recorded.  To save time and simplify the code, this is recorded
+    // as the ID of the error message that would be generated if the I/O failed.
+    // This means that we must make sure that all possible "origins" take the
+    // same arguments in their message in the same order.
+    isc::log::MessageID         origin;     ///< Origin of last asynchronous I/O
+    uint8_t                     staging[IOFetch::STAGING_LENGTH];
+                                            ///< Temporary array for received data
+
+    /// \brief Constructor
+    ///
+    /// Just fills in the data members of the IOFetchData structure
+    ///
+    /// \param proto Either IOFetch::TCP or IOFetch::UDP.
+    /// \param service I/O Service object to handle the asynchronous
+    ///        operations.
+    /// \param query DNS question to send to the upstream server.
+    /// \param address IP address of upstream server
+    /// \param port Port to use for the query
+    /// \param buff Output buffer into which the response (in wire format)
+    ///        is written (if a response is received).
+    /// \param cb Callback object containing the callback to be called
+    ///        when we terminate.  The caller is responsible for managing this
+    ///        object and deleting it if necessary.
+    /// \param wait Timeout for the fetch (in ms).
+    ///
+    /// TODO: May need to alter constructor (see comment 4 in Trac ticket #554)
+    IOFetchData(IOFetch::Protocol proto, IOService& service,
+        const isc::dns::Question& query, const IOAddress& address,
+        uint16_t port, isc::dns::OutputBufferPtr& buff, IOFetch::Callback* cb,
+        int wait)
+        :
+        socket((proto == IOFetch::UDP) ?
+            static_cast<IOAsioSocket<IOFetch>*>(
+                new UDPSocket<IOFetch>(service)) :
+            static_cast<IOAsioSocket<IOFetch>*>(
+                new TCPSocket<IOFetch>(service))
+            ),
+        remote((proto == IOFetch::UDP) ?
+            static_cast<IOEndpoint*>(new UDPEndpoint(address, port)) :
+            static_cast<IOEndpoint*>(new TCPEndpoint(address, port))
+            ),
+        question(query),
+        msgbuf(new isc::dns::OutputBuffer(512)),
+        received(buff),
+
+        callback(cb),
+        timer(service.get_io_service()),
+        protocol(proto),
+        cumulative(0),
+        expected(0),
+        offset(0),
+        stopped(false),
+        timeout(wait),
+        origin(ASIO_UNKORIGIN),
+        staging()
+    {}
+};
 
-namespace asiolink {
 /// IOFetch Constructor - just initialize the private data
 
-IOFetch::IOFetch(int protocol, IOService& service,
+IOFetch::IOFetch(Protocol protocol, IOService& service,
     const isc::dns::Question& question, const IOAddress& address, uint16_t port,
-    isc::dns::OutputBufferPtr& buff, Callback* cb, int wait)
+    OutputBufferPtr& buff, Callback* cb, int wait)
     :
-    data_(new IOFetch::IOFetchData(protocol, service, question, address,
+    data_(new IOFetchData(protocol, service, question, address,
         port, buff, cb, wait))
 {
 }
 
+// Return protocol in use.
+
+IOFetch::Protocol
+IOFetch::getProtocol() const {
+    return (data_->protocol);
+}
+
 /// The function operator is implemented with the "stackless coroutine"
 /// pattern; see internal/coroutine.h for details.
 
 void
-IOFetch::operator()(error_code ec, size_t length) {
-    if (ec || data_->stopped) {
+IOFetch::operator()(asio::error_code ec, size_t length) {
+
+    if (data_->stopped) {
+        return;
+    } else if (ec) {
+        logIOFailure(ec);
         return;
     }
 
@@ -66,7 +180,6 @@ IOFetch::operator()(error_code ec, size_t length) {
         /// declarations.
         {
             Message msg(Message::RENDER);
-            
             msg.setQid(QidGenerator::getInstance().generateQid());
             msg.setOpcode(Opcode::QUERY());
             msg.setRcode(Rcode::NOERROR());
@@ -74,17 +187,10 @@ IOFetch::operator()(error_code ec, size_t length) {
             msg.addQuestion(data_->question);
             MessageRenderer renderer(*data_->msgbuf);
             msg.toWire(renderer);
-
-            // As this is a new fetch, clear the amount of data received
-            data_->cumulative = 0;
-
-            dlog("Sending " + msg.toText() + " to " +
-                data_->remote->getAddress().toText());
         }
 
-
-        // If we timeout, we stop, which will shutdown everything and
-        // cancel all other attempts to run inside the coroutine
+        // If we timeout, we stop, which will can cancel outstanding I/Os and
+        // shutdown everything.
         if (data_->timeout != -1) {
             data_->timer.expires_from_now(boost::posix_time::milliseconds(
                 data_->timeout));
@@ -93,13 +199,17 @@ IOFetch::operator()(error_code ec, size_t length) {
         }
 
         // Open a connection to the target system.  For speed, if the operation
-        // was completed synchronously (i.e. UDP operation) we bypass the yield.
-        if (data_->socket->open(data_->remote.get(), *this)) {
-            CORO_YIELD;
+        // is synchronous (i.e. UDP operation) we bypass the yield.
+        data_->origin = ASIO_OPENSOCK;
+        if (data_->socket->isOpenSynchronous()) {
+            data_->socket->open(data_->remote.get(), *this);
+        } else {
+            CORO_YIELD data_->socket->open(data_->remote.get(), *this);
         }
 
-        // Begin an asynchronous send, and then yield.  When the send completes
-        // send completes, we will resume immediately after this point.
+        // Begin an asynchronous send, and then yield.  When the send completes,
+        // we will resume immediately after this point.
+        data_->origin = ASIO_SENDSOCK;
         CORO_YIELD data_->socket->asyncSend(data_->msgbuf->getData(),
             data_->msgbuf->getLength(), data_->remote.get(), *this);
 
@@ -110,24 +220,33 @@ IOFetch::operator()(error_code ec, size_t length) {
         // we need to yield ... and we *really* don't want to set up another
         // coroutine within that method.)  So after each receive (and yield),
         // we check if the operation is complete and if not, loop to read again.
+        //
+        // Another concession to TCP is that the amount of is contained in the
+        // first two bytes.  This leads to two problems:
+        //
+        // a) We don't want those bytes in the return buffer.
+        // b) They may not both arrive in the first I/O.
+        //
+        // So... we need to loop until we have at least two bytes, then store
+        // the expected amount of data.  Then we need to loop until we have
+        // received all the data before copying it back to the user's buffer.
+        // And we want to minimise the amount of copying...
+
+        data_->origin = ASIO_RECVSOCK;
+        data_->cumulative = 0;          // No data yet received
+        data_->offset = 0;              // First data into start of buffer
         do {
-            CORO_YIELD data_->socket->asyncReceive(data_->data.get(),
-                static_cast<size_t>(MAX_LENGTH), data_->cumulative,
-                data_->remote.get(), *this);
-        } while (!data_->socket->receiveComplete(data_->data.get(), length,
-            data_->cumulative));
-
-        // The message is not rendered yet, so we can't print it easily
-        dlog("Received response from " + data_->remote->getAddress().toText());
-
-        /// Copy the answer into the response buffer.  (TODO: If the
-        /// OutputBuffer object were made to meet the requirements of
-        /// a MutableBufferSequence, then it could be written to directly
-        /// by async_receive_from() and this additional copy step would
-        /// be unnecessary.)
-        data_->buffer->writeData(data_->data.get(), length);
-
-        // Finished with this socket, so close it.
+            CORO_YIELD data_->socket->asyncReceive(data_->staging,
+                                                   static_cast<size_t>(STAGING_LENGTH),
+                                                   data_->offset,
+                                                   data_->remote.get(), *this);
+        } while (!data_->socket->processReceivedData(data_->staging, length,
+                                                     data_->cumulative, data_->offset,
+                                                     data_->expected, data_->received));
+
+        // Finished with this socket, so close it.  This will not generate an
+        // I/O error, but reset the origin to unknown in case we change this.
+        data_->origin = ASIO_UNKORIGIN;
         data_->socket->close();
 
         /// We are done
@@ -139,9 +258,8 @@ IOFetch::operator()(error_code ec, size_t length) {
 // query finishes or when the timer times out.  Either way, it sets the
 // "stopped_" flag and cancels anything that is in progress.
 //
-// As the function may be entered multiple times as things wind down, the
-// stopped_ flag checks if stop() has already been called.  If it has,
-// subsequent calls are no-ops.
+// As the function may be entered multiple times as things wind down, it checks
+// if the stopped_ flag is already set.  If it is, the call is a no-op.
 
 void
 IOFetch::stop(Result result) {
@@ -158,20 +276,46 @@ IOFetch::stop(Result result) {
         // variable should be done inside a mutex (and the stopped_ variable
         // declared as "volatile").
         //
+        // The numeric arguments indicate the debug level, with the lower
+        // numbers indicating the most important information.  The relative
+        // values are somewhat arbitrary.
+        //
+        // Although Logger::debug checks the debug flag internally, doing it
+        // below before calling Logger::debug avoids the overhead of a string
+        // conversion in the common case when debug is not enabled.
+        //
         // TODO: Update testing of stopped_ if threads are used.
         data_->stopped = true;
-
         switch (result) {
             case TIME_OUT:
-                dlog("Query timed out");
+                if (logger.isDebugEnabled(1)) {
+                    logger.debug(20, ASIO_RECVTMO,
+                                 data_->remote->getAddress().toText().c_str(),
+                                 static_cast<int>(data_->remote->getPort()));
+                }
+                break;
+
+            case SUCCESS:
+                if (logger.isDebugEnabled(50)) {
+                    logger.debug(30, ASIO_FETCHCOMP,
+                                 data_->remote->getAddress().toText().c_str(),
+                                 static_cast<int>(data_->remote->getPort()));
+                }
                 break;
 
             case STOPPED:
-                dlog("Query stopped");
+                // Fetch has been stopped for some other reason.  This is
+                // allowed but as it is unusual it is logged, but with a lower
+                // debug level than a timeout (which is totally normal).
+                logger.debug(1, ASIO_FETCHSTOP,
+                             data_->remote->getAddress().toText().c_str(),
+                             static_cast<int>(data_->remote->getPort()));
                 break;
 
             default:
-                ;
+                logger.error(ASIO_UNKRESULT, static_cast<int>(result),
+                             data_->remote->getAddress().toText().c_str(),
+                             static_cast<int>(data_->remote->getPort()));
         }
 
         // Stop requested, cancel and I/O's on the socket and shut it down,
@@ -185,10 +329,26 @@ IOFetch::stop(Result result) {
         if (data_->callback) {
             (*(data_->callback))(result);
         }
+    }
+}
 
-        // Mark that stop() has now been called.
+// Log an error - called on I/O failure
 
-    }
+void IOFetch::logIOFailure(asio::error_code ec) {
+
+    // Should only get here with a known error code.
+    assert((data_->origin == ASIO_OPENSOCK) ||
+           (data_->origin == ASIO_SENDSOCK) ||
+           (data_->origin == ASIO_RECVSOCK) ||
+           (data_->origin == ASIO_UNKORIGIN));
+
+    static const char* PROTOCOL[2] = {"TCP", "UDP"};
+    logger.error(data_->origin,
+                 ec.value(),
+                 ((data_->remote->getProtocol() == IPPROTO_TCP) ?
+                     PROTOCOL[0] : PROTOCOL[1]),
+                 data_->remote->getAddress().toText().c_str(),
+                 static_cast<int>(data_->remote->getPort()));
 }
 
 } // namespace asiolink

src/lib/asiolink/io_fetch.h

@@ -17,31 +17,23 @@
 
 #include <config.h>
 
-#include <netinet/in.h>
-#include <sys/socket.h>
-#include <unistd.h>             // for some IPC/network system calls
-
 #include <boost/shared_array.hpp>
 #include <boost/shared_ptr.hpp>
 #include <boost/date_time/posix_time/posix_time_types.hpp>
-#include <asio/deadline_timer.hpp>
 
 #include <coroutine.h>
 
+#include <asio/error_code.hpp>
+
 #include <dns/buffer.h>
 #include <dns/question.h>
 
-#include <asiolink/io_asio_socket.h>
-#include <asiolink/io_endpoint.h>
-#include <asiolink/io_service.h>
-#include <asiolink/tcp_socket.h>
-#include <asiolink/tcp_endpoint.h>
-#include <asiolink/udp_socket.h>
-#include <asiolink/udp_endpoint.h>
-
-
 namespace asiolink {
 
+// Forward declarations
+class IOAddress;
+class IOFetchData;
+class IOService;
 
 /// \brief Upstream Fetch Processing
 ///
@@ -51,6 +43,23 @@ namespace asiolink {
 
 class IOFetch : public coroutine {
 public:
+    /// \brief Protocol to use on the fetch
+    enum Protocol {
+        UDP = 0,
+        TCP = 1
+    };
+
+    /// \brief Origin of Asynchronous I/O Call
+    ///
+    /// Indicates what initiated an asynchronous I/O call and used in deciding
+    /// what error message to output if the I/O fails.
+    enum Origin {
+        NONE = 0,           ///< No asynchronous call outstanding
+        OPEN = 1,
+        SEND = 2,
+        RECEIVE = 3,
+        CLOSE = 4
+    };
 
     /// \brief Result of Upstream Fetch
     ///
@@ -59,9 +68,9 @@ public:
     /// even if the contents of the packet indicate that some error occurred.
     enum Result {
         SUCCESS = 0,        ///< Success, fetch completed
-        TIME_OUT,           ///< Failure, fetch timed out
-        STOPPED,            ///< Control code, fetch has been stopped
-        NOTSET              ///< For testing, indicates value not set
+        TIME_OUT = 1,       ///< Failure, fetch timed out
+        STOPPED = 2,        ///< Control code, fetch has been stopped
+        NOTSET = 3          ///< For testing, indicates value not set
     };
 
     // The next enum is a "trick" to allow constants to be defined in a class
@@ -69,7 +78,7 @@ public:
 
     /// \brief Integer Constants
     enum {
-        MAX_LENGTH = 4096   ///< Maximum size of receive buffer
+        STAGING_LENGTH = 8192   ///< Size of staging buffer
     };
 
     /// \brief I/O Fetch Callback
@@ -95,82 +104,12 @@ public:
         virtual ~Callback()
         {}
 
-        /// \brief Callback method called when the fetch completes
-        ///
-        /// \brief result Result of the fetch
-        virtual void operator()(Result result) = 0;
-    };
-
-    /// \brief IOFetch Data
-    ///
-    /// The data for IOFetch is held in a separate struct pointed to by a
-    /// shared_ptr object.  This is because the IOFetch object will be copied
-    /// often (it is used as a coroutine and passed as callback to many
-    /// async_*() functions) and we want keep the same data).  Organising the
-    /// data in this way keeps copying to a minimum.
-    struct IOFetchData {
-
-        // The next two members are shared pointers to a base class because what
-        // is actually instantiated depends on whether the fetch is over UDP or
-        // TCP, which is not known until construction of the IOFetch.  Use of
-        // a shared pointer here is merely to ensure deletion when the data
-        // object is deleted.
-        boost::shared_ptr<IOAsioSocket<IOFetch> > socket;
-                                                ///< Socket to use for I/O
-        boost::shared_ptr<IOEndpoint> remote;   ///< Where the fetch was sent
-        isc::dns::Question          question;   ///< Question to be asked
-        isc::dns::OutputBufferPtr   msgbuf;     ///< Wire buffer for question
-        isc::dns::OutputBufferPtr   buffer;     ///< Received data held here
-        boost::shared_array<char>   data;       ///< Temporary array for data
-        IOFetch::Callback*          callback;   ///< Called on I/O Completion
-        size_t                      cumulative; ///< Cumulative received amount
-        bool                        stopped;    ///< Have we stopped running?
-        asio::deadline_timer        timer;      ///< Timer to measure timeouts
-        int                         timeout;    ///< Timeout in ms
-
-        /// \brief Constructor
-        ///
-        /// Just fills in the data members of the IOFetchData structure
+        /// \brief Callback method
         ///
-        /// \param protocol either IPPROTO_UDP or IPPROTO_TCP
-        /// \param service I/O Service object to handle the asynchronous
-        ///     operations.
-        /// \param query DNS question to send to the upstream server.
-        /// \param address IP address of upstream server
-        /// \param port Port to use for the query
-        /// \param buff Output buffer into which the response (in wire format)
-        ///     is written (if a response is received).
-        /// \param cb Callback object containing the callback to be called
-        ///     when we terminate.  The caller is responsible for managing this
-        ///     object and deleting it if necessary.
-        /// \param wait Timeout for the fetch (in ms).
+        /// This is the method called when the fetch completes.
         ///
-        /// TODO: May need to alter constructor (see comment 4 in Trac ticket #554)
-        IOFetchData(int protocol, IOService& service,
-            const isc::dns::Question& query, const IOAddress& address,
-            uint16_t port, isc::dns::OutputBufferPtr& buff, Callback* cb,
-            int wait)
-            :
-            socket((protocol == IPPROTO_UDP) ?
-                static_cast<IOAsioSocket<IOFetch>*>(
-                    new UDPSocket<IOFetch>(service)) :
-                static_cast<IOAsioSocket<IOFetch>*>(
-                    new TCPSocket<IOFetch>(service))
-                ),
-            remote((protocol == IPPROTO_UDP) ?
-                static_cast<IOEndpoint*>(new UDPEndpoint(address, port)) :
-                static_cast<IOEndpoint*>(new TCPEndpoint(address, port))
-                ),
-            question(query),
-            msgbuf(new isc::dns::OutputBuffer(512)),
-            buffer(buff),
-            data(new char[IOFetch::MAX_LENGTH]),
-            callback(cb),
-            cumulative(0),
-            stopped(false),
-            timer(service.get_io_service()),
-            timeout(wait)
-        {}
+        /// \param result Result of the fetch
+        virtual void operator()(Result result) = 0;
     };
 
     /// \brief Constructor.
@@ -179,7 +118,7 @@ public:
     ///
     /// TODO: Need to randomise the source port
     ///
-    /// \param protocol Fetch protocol, either IPPROTO_UDP or IPPROTO_TCP
+    /// \param protocol Fetch protocol, either IOFetch::TCP or IOFetch::UDP
     /// \param service I/O Service object to handle the asynchronous
     ///     operations.
     /// \param question DNS question to send to the upstream server.
@@ -193,11 +132,16 @@ public:
     /// (default = 53)
     /// \param wait Timeout for the fetch (in ms).  The default value of
     ///     -1 indicates no timeout.
-    IOFetch(int protocol, IOService& service,
+    IOFetch(Protocol protocol, IOService& service,
         const isc::dns::Question& question, const IOAddress& address,
         uint16_t port, isc::dns::OutputBufferPtr& buff, Callback* cb,
         int wait = -1);
-    
+
+    /// \brief Return Current Protocol
+    ///
+    /// \return Protocol associated with this IOFetch object.
+    Protocol getProtocol() const;
+
     /// \brief Coroutine entry point
     ///
     /// The operator() method is the method in which the coroutine code enters
@@ -205,8 +149,7 @@ public:
     ///
     /// \param ec Error code, the result of the last asynchronous I/O operation.
     /// \param length Amount of data received on the last asynchronous read
-    void operator()(asio::error_code ec = asio::error_code(),
-        size_t length = 0);
+    void operator()(asio::error_code ec = asio::error_code(), size_t length = 0);
 
     /// \brief Terminate query
     ///
@@ -217,6 +160,16 @@ public:
     void stop(Result reason = STOPPED);
 
 private:
+    /// \brief Log I/O Failure
+    ///
+    /// Records an I/O failure to the log file
+    ///
+    /// \param ec ASIO error code
+    void logIOFailure(asio::error_code ec);
+
+    // Member variables.  All data is in a structure pointed to by a shared
+    // pointer.  The IOFetch object is copied a number of times during its
+    // life, and only requiring a pointer to be copied reduces overhead.
     boost::shared_ptr<IOFetchData>  data_;   ///< Private data
 
 };

src/lib/asiolink/io_message.h

@@ -98,7 +98,3 @@ private:
 
 }      // asiolink
 #endif // __IO_MESSAGE_H
-
-// Local Variables: 
-// mode: c++
-// End: 

src/lib/asiolink/recursive_query.cc

@@ -1,543 +0,0 @@
-// Copyright (C) 2011  Internet Systems Consortium, Inc. ("ISC")
-//
-// Permission to use, copy, modify, and/or distribute this software for any
-// purpose with or without fee is hereby granted, provided that the above
-// copyright notice and this permission notice appear in all copies.
-//
-// THE SOFTWARE IS PROVIDED "AS IS" AND ISC DISCLAIMS ALL WARRANTIES WITH
-// REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
-// AND FITNESS.  IN NO EVENT SHALL ISC BE LIABLE FOR ANY SPECIAL, DIRECT,
-// INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
-// LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
-// OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
-// PERFORMANCE OF THIS SOFTWARE.
-
-#include <netinet/in.h>
-#include <stdlib.h>
-#include <sys/socket.h>
-#include <unistd.h>             // for some IPC/network system calls
-
-#include <boost/lexical_cast.hpp>
-#include <boost/bind.hpp>
-
-#include <config.h>
-
-#include <log/dummylog.h>
-
-#include <dns/question.h>
-#include <dns/message.h>
-#include <dns/opcode.h>
-
-#include <resolve/resolve.h>
-#include <cache/resolver_cache.h>
-
-#include <asio.hpp>
-#include <asiolink/dns_service.h>
-#include <asiolink/io_fetch.h>
-#include <asiolink/io_service.h>
-#include <asiolink/recursive_query.h>
-
-using isc::log::dlog;
-using namespace isc::dns;
-
-namespace asiolink {
-
-typedef std::vector<std::pair<std::string, uint16_t> > AddressVector;
-
-// Here we do not use the typedef above, as the SunStudio compiler
-// mishandles this in its name mangling, and wouldn't compile.
-// We can probably use a typedef, but need to move it to a central
-// location and use it consistently.
-RecursiveQuery::RecursiveQuery(DNSService& dns_service,
-    const std::vector<std::pair<std::string, uint16_t> >& upstream,
-    const std::vector<std::pair<std::string, uint16_t> >& upstream_root,
-    int query_timeout, int client_timeout, int lookup_timeout,
-    unsigned retries) :
-    dns_service_(dns_service), upstream_(new AddressVector(upstream)),
-    upstream_root_(new AddressVector(upstream_root)),
-    query_timeout_(query_timeout), client_timeout_(client_timeout),
-    lookup_timeout_(lookup_timeout), retries_(retries)
-{}
-
-namespace {
-
-typedef std::pair<std::string, uint16_t> addr_t;
-
-/*
- * This is a query in progress. When a new query is made, this one holds
- * the context information about it, like how many times we are allowed
- * to retry on failure, what to do when we succeed, etc.
- *
- * Used by RecursiveQuery::sendQuery.
- */
-class RunningQuery : public IOFetch::Callback {
-private:
-    // The io service to handle async calls
-    IOService& io_;
-
-    // Info for (re)sending the query (the question and destination)
-    Question question_;
-
-    // This is where we build and store our final answer
-    MessagePtr answer_message_;
-
-    // currently we use upstream as the current list of NS records
-    // we should differentiate between forwarding and resolving
-    boost::shared_ptr<AddressVector> upstream_;
-
-    // root servers...just copied over to the zone_servers_
-    boost::shared_ptr<AddressVector> upstream_root_;
-
-    // Buffer to store the result.
-    OutputBufferPtr buffer_;
-
-    // Server to notify when we succeed or fail
-    //shared_ptr<DNSServer> server_;
-    isc::resolve::ResolverInterface::CallbackPtr resolvercallback_;
-
-    // To prevent both unreasonably long cname chains and cname loops,
-    // we simply keep a counter of the number of CNAMEs we have
-    // followed so far (and error if it exceeds RESOLVER_MAX_CNAME_CHAIN
-    // from lib/resolve/response_classifier.h)
-    unsigned cname_count_;
-
-    /*
-     * TODO Do something more clever with timeouts. In the long term, some
-     *     computation of average RTT, increase with each retry, etc.
-     */
-    // Timeout information
-    int query_timeout_;
-    unsigned retries_;
-
-    // normal query state
-
-    // Not using NSAS at this moment, so we keep a list
-    // of 'current' zone servers
-    std::vector<addr_t> zone_servers_;
-
-    // Update the question that will be sent to the server
-    void setQuestion(const Question& new_question) {
-        question_ = new_question;
-    }
-
-    // TODO: replace by our wrapper
-    asio::deadline_timer client_timer;
-    asio::deadline_timer lookup_timer;
-
-    size_t queries_out_;
-    
-    // If we timed out ourselves (lookup timeout), stop issuing queries
-    bool done_;
-
-    // If we have a client timeout, we send back an answer, but don't
-    // stop. We use this variable to make sure we don't send another
-    // answer if we do find one later (or if we have a lookup_timeout)
-    bool answer_sent_;
-
-    // Reference to our cache
-    isc::cache::ResolverCache& cache_;
-
-    // perform a single lookup; first we check the cache to see
-    // if we have a response for our query stored already. if
-    // so, call handlerecursiveresponse(), if not, we call send()
-    void doLookup() {
-        dlog("doLookup: try cache");
-        Message cached_message(Message::RENDER);
-        isc::resolve::initResponseMessage(question_, cached_message);
-        if (cache_.lookup(question_.getName(), question_.getType(),
-                          question_.getClass(), cached_message)) {
-            dlog("Message found in cache, returning that");
-            handleRecursiveAnswer(cached_message);
-        } else {
-            send();
-        }
-        
-    }
-
-    // (re)send the query to the server.
-    void send() {
-        const int uc = upstream_->size();
-        const int zs = zone_servers_.size();
-        buffer_->clear();
-        if (uc > 0) {
-            int serverIndex = rand() % uc;
-            dlog("Sending upstream query (" + question_.toText() +
-                ") to " + upstream_->at(serverIndex).first);
-            IOFetch query(IPPROTO_UDP, io_, question_,
-                upstream_->at(serverIndex).first,
-                upstream_->at(serverIndex).second, buffer_, this,
-                query_timeout_);
-            ++queries_out_;
-            io_.get_io_service().post(query);
-        } else if (zs > 0) {
-            int serverIndex = rand() % zs;
-            dlog("Sending query to zone server (" + question_.toText() +
-                ") to " + zone_servers_.at(serverIndex).first);
-            IOFetch query(IPPROTO_UDP, io_, question_,
-                zone_servers_.at(serverIndex).first,
-                zone_servers_.at(serverIndex).second, buffer_, this,
-                query_timeout_);
-            ++queries_out_;
-            io_.get_io_service().post(query);
-        } else {
-            dlog("Error, no upstream servers to send to.");
-        }
-    }
-    
-    // This function is called by operator() if there is an actual
-    // answer from a server and we are in recursive mode
-    // depending on the contents, we go on recursing or return
-    //
-    // Note that the footprint may change as this function may
-    // need to append data to the answer we are building later.
-    //
-    // returns true if we are done (either we have an answer or an
-    //              error message)
-    // returns false if we are not done
-    bool handleRecursiveAnswer(const Message& incoming) {
-        dlog("Handle response");
-        // In case we get a CNAME, we store the target
-        // here (classify() will set it when it walks through
-        // the cname chain to verify it).
-        Name cname_target(question_.getName());
-        
-        isc::resolve::ResponseClassifier::Category category =
-            isc::resolve::ResponseClassifier::classify(
-                question_, incoming, cname_target, cname_count_, true);
-
-        bool found_ns_address = false;
-            
-        // If the packet is OK, store it in the cache
-        if (!isc::resolve::ResponseClassifier::error(category)) {
-            cache_.update(incoming);
-        }
-
-        switch (category) {
-        case isc::resolve::ResponseClassifier::ANSWER:
-        case isc::resolve::ResponseClassifier::ANSWERCNAME:
-            // Done. copy and return.
-            isc::resolve::copyResponseMessage(incoming, answer_message_);
-            return true;
-            break;
-        case isc::resolve::ResponseClassifier::CNAME:
-            dlog("Response is CNAME!");
-            // (unfinished) CNAME. We set our question_ to the CNAME
-            // target, then start over at the beginning (for now, that
-            // is, we reset our 'current servers' to the root servers).
-            if (cname_count_ >= RESOLVER_MAX_CNAME_CHAIN) {
-                // just give up
-                dlog("CNAME chain too long");
-                isc::resolve::makeErrorMessage(answer_message_,
-                                               Rcode::SERVFAIL());
-                return true;
-            }
-
-            answer_message_->appendSection(Message::SECTION_ANSWER,
-                                           incoming);
-            setZoneServersToRoot();
-
-            question_ = Question(cname_target, question_.getClass(),
-                                 question_.getType());
-
-            dlog("Following CNAME chain to " + question_.toText());
-            doLookup();
-            return false;
-            break;
-        case isc::resolve::ResponseClassifier::NXDOMAIN:
-            // NXDOMAIN, just copy and return.
-            isc::resolve::copyResponseMessage(incoming, answer_message_);
-            return true;
-            break;
-        case isc::resolve::ResponseClassifier::REFERRAL:
-            // Referral. For now we just take the first glue address
-            // we find and continue with that
-            zone_servers_.clear();
-
-            for (RRsetIterator rrsi = incoming.beginSection(Message::SECTION_ADDITIONAL);
-                 rrsi != incoming.endSection(Message::SECTION_ADDITIONAL) && !found_ns_address;
-                 rrsi++) {
-                ConstRRsetPtr rrs = *rrsi;
-                if (rrs->getType() == RRType::A()) {
-                    // found address
-                    RdataIteratorPtr rdi = rrs->getRdataIterator();
-                    // just use the first for now
-                    if (!rdi->isLast()) {
-                        std::string addr_str = rdi->getCurrent().toText();
-                        dlog("[XX] first address found: " + addr_str);
-                        // now we have one address, simply
-                        // resend that exact same query
-                        // to that address and yield, when it
-                        // returns, loop again.
-                        
-                        // TODO should use NSAS
-                        zone_servers_.push_back(addr_t(addr_str, 53));
-                        found_ns_address = true;
-                        break;
-                    }
-                }
-            }
-            if (found_ns_address) {
-                // next resolver round
-                // we do NOT use doLookup() here, but send() (i.e. we
-                // skip the cache), since if we had the final answer
-                // instead of a delegation cached, we would have been
-                // there by now.
-                send();
-                return false;
-            } else {
-                dlog("[XX] no ready-made addresses in additional. need nsas.");
-                // TODO this will result in answering with the delegation. oh well
-                isc::resolve::copyResponseMessage(incoming, answer_message_);
-                return true;
-            }
-            break;
-        case isc::resolve::ResponseClassifier::EMPTY:
-        case isc::resolve::ResponseClassifier::EXTRADATA:
-        case isc::resolve::ResponseClassifier::INVNAMCLASS:
-        case isc::resolve::ResponseClassifier::INVTYPE:
-        case isc::resolve::ResponseClassifier::MISMATQUEST:
-        case isc::resolve::ResponseClassifier::MULTICLASS:
-        case isc::resolve::ResponseClassifier::NOTONEQUEST:
-        case isc::resolve::ResponseClassifier::NOTRESPONSE:
-        case isc::resolve::ResponseClassifier::NOTSINGLE:
-        case isc::resolve::ResponseClassifier::OPCODE:
-        case isc::resolve::ResponseClassifier::RCODE:
-        case isc::resolve::ResponseClassifier::TRUNCATED:
-            // Should we try a different server rather than SERVFAIL?
-            isc::resolve::makeErrorMessage(answer_message_,
-                                           Rcode::SERVFAIL());
-            return true;
-            break;
-        }
-        // should not be reached. assert here?
-        dlog("[FATAL] unreachable code");
-        return true;
-    }
-    
-public:
-    RunningQuery(IOService& io,
-        const Question &question,
-        MessagePtr answer_message,
-        boost::shared_ptr<AddressVector> upstream,
-        boost::shared_ptr<AddressVector> upstream_root,
-        OutputBufferPtr buffer,
-        isc::resolve::ResolverInterface::CallbackPtr cb,
-        int query_timeout, int client_timeout, int lookup_timeout,
-        unsigned retries,
-        isc::cache::ResolverCache& cache) :
-        io_(io),
-        question_(question),
-        answer_message_(answer_message),
-        upstream_(upstream),
-        upstream_root_(upstream_root),
-        buffer_(buffer),
-        resolvercallback_(cb),
-        cname_count_(0),
-        query_timeout_(query_timeout),
-        retries_(retries),
-        client_timer(io.get_io_service()),
-        lookup_timer(io.get_io_service()),
-        queries_out_(0),
-        done_(false),
-        answer_sent_(false),
-        cache_(cache)
-    {
-        // Setup the timer to stop trying (lookup_timeout)
-        if (lookup_timeout >= 0) {
-            lookup_timer.expires_from_now(
-                boost::posix_time::milliseconds(lookup_timeout));
-            lookup_timer.async_wait(boost::bind(&RunningQuery::stop, this, false));
-        }
-        
-        // Setup the timer to send an answer (client_timeout)
-        if (client_timeout >= 0) {
-            client_timer.expires_from_now(
-                boost::posix_time::milliseconds(client_timeout));
-            client_timer.async_wait(boost::bind(&RunningQuery::clientTimeout, this));
-        }
-        
-        // should use NSAS for root servers
-        // Adding root servers if not a forwarder
-        if (upstream_->empty()) {
-            setZoneServersToRoot();
-        }
-
-        doLookup();
-    }
-
-    void setZoneServersToRoot() {
-        zone_servers_.clear();
-        if (upstream_root_->empty()) { //if no root ips given, use this
-            zone_servers_.push_back(addr_t("192.5.5.241", 53));
-        } else {
-            // copy the list
-            dlog("Size is " + 
-                boost::lexical_cast<std::string>(upstream_root_->size()) + 
-                "\n");
-            for(AddressVector::iterator it = upstream_root_->begin();
-                it < upstream_root_->end(); ++it) {
-            zone_servers_.push_back(addr_t(it->first,it->second));
-            dlog("Put " + zone_servers_.back().first + "into root list\n");
-            }
-        }
-    }
-    virtual void clientTimeout() {
-        // Return a SERVFAIL, but do not stop until
-        // we have an answer or timeout ourselves
-        isc::resolve::makeErrorMessage(answer_message_,
-                                       Rcode::SERVFAIL());
-        if (!answer_sent_) {
-            answer_sent_ = true;
-            resolvercallback_->success(answer_message_);
-        }
-    }
-
-    virtual void stop(bool resume) {
-        // if we cancel our timers, we will still get an event for
-        // that, so we cannot delete ourselves just yet (those events
-        // would be bound to a deleted object)
-        // cancel them one by one, both cancels should get us back
-        // here again.
-        // same goes if we have an outstanding query (can't delete
-        // until that one comes back to us)
-        done_ = true;
-        if (resume && !answer_sent_) {
-            answer_sent_ = true;
-
-            // There are two types of messages we could store in the
-            // cache;
-            // 1. answers to our fetches from authoritative servers,
-            //    exactly as we receive them, and
-            // 2. answers to queries we received from clients, which
-            //    have received additional processing (following CNAME
-            //    chains, for instance)
-            //
-            // Doing only the first would mean we would have to re-do
-            // processing when we get data from our cache, and doing
-            // only the second would miss out on the side-effect of
-            // having nameserver data in our cache.
-            //
-            // So right now we do both. Since the cache (currently)
-            // stores Messages on their question section only, this
-            // does mean that we overwrite the messages we stored in
-            // the previous iteration if we are following a delegation.
-            cache_.update(*answer_message_);
-
-            resolvercallback_->success(answer_message_);
-        } else {
-            resolvercallback_->failure();
-        }
-        if (lookup_timer.cancel() != 0) {
-            return;
-        }
-        if (client_timer.cancel() != 0) {
-            return;
-        }
-        if (queries_out_ > 0) {
-            return;
-        }
-        delete this;
-    }
-
-    // This function is used as callback from DNSQuery.
-    virtual void operator()(IOFetch::Result result) {
-        // XXX is this the place for TCP retry?
-        --queries_out_;
-        if (!done_ && result != IOFetch::TIME_OUT) {
-            // we got an answer
-            Message incoming(Message::PARSE);
-            InputBuffer ibuf(buffer_->getData(), buffer_->getLength());
-            incoming.fromWire(ibuf);
-
-            if (upstream_->size() == 0 &&
-                incoming.getRcode() == Rcode::NOERROR()) {
-                done_ = handleRecursiveAnswer(incoming);
-            } else {
-                isc::resolve::copyResponseMessage(incoming, answer_message_);
-                done_ = true;
-            }
-            
-            if (done_) {
-                stop(true);
-            }
-        } else if (!done_ && retries_--) {
-            // We timed out, but we have some retries, so send again
-            dlog("Timeout, resending query");
-            send();
-        } else {
-            // out of retries, give up for now
-            stop(false);
-        }
-    }
-};
-
-}
-
-void
-RecursiveQuery::resolve(const QuestionPtr& question,
-    const isc::resolve::ResolverInterface::CallbackPtr callback)
-{
-    IOService& io = dns_service_.getIOService();
-
-    MessagePtr answer_message(new Message(Message::RENDER));
-    isc::resolve::initResponseMessage(*question, *answer_message);
-
-    OutputBufferPtr buffer(new OutputBuffer(0));
-
-    dlog("Try out cache first (direct call to resolve)");
-    // First try to see if we have something cached in the messagecache
-    if (cache_.lookup(question->getName(), question->getType(),
-                      question->getClass(), *answer_message)) {
-        dlog("Message found in cache, returning that");
-        // TODO: err, should cache set rcode as well?
-        answer_message->setRcode(Rcode::NOERROR());
-        callback->success(answer_message);
-    } else {
-        dlog("Message not found in cache, starting recursive query");
-        // It will delete itself when it is done
-        new RunningQuery(io, *question, answer_message, upstream_,
-                         upstream_root_, buffer, callback, query_timeout_,
-                         client_timeout_, lookup_timeout_, retries_,
-                         cache_);
-    }
-}
-
-void
-RecursiveQuery::resolve(const Question& question,
-                        MessagePtr answer_message,
-                        OutputBufferPtr buffer,
-                        DNSServer* server)
-{
-    // XXX: eventually we will need to be able to determine whether
-    // the message should be sent via TCP or UDP, or sent initially via
-    // UDP and then fall back to TCP on failure, but for the moment
-    // we're only going to handle UDP.
-    IOService& io = dns_service_.getIOService();
-
-    isc::resolve::ResolverInterface::CallbackPtr crs(
-        new isc::resolve::ResolverCallbackServer(server));
-
-    // TODO: general 'prepareinitialanswer'
-    answer_message->setOpcode(isc::dns::Opcode::QUERY());
-    answer_message->addQuestion(question);
-    
-    // First try to see if we have something cached in the messagecache
-    dlog("Try out cache first (started by incoming event)");
-    if (cache_.lookup(question.getName(), question.getType(),
-                      question.getClass(), *answer_message)) {
-        dlog("Message found in cache, returning that");
-        // TODO: err, should cache set rcode as well?
-        answer_message->setRcode(Rcode::NOERROR());
-        crs->success(answer_message);
-    } else {
-        dlog("Message not found in cache, starting recursive query");
-        // It will delete itself when it is done
-        new RunningQuery(io, question, answer_message, upstream_, upstream_root_,
-                             buffer, crs, query_timeout_, client_timeout_,
-                             lookup_timeout_, retries_, cache_);
-    }
-}
-
-
-
-} // namespace asiolink

src/lib/asiolink/tcp_endpoint.h

@@ -24,32 +24,33 @@
 namespace asiolink {
 
 /// \brief The \c TCPEndpoint class is a concrete derived class of
-/// \c IOEndpoint that represents an endpoint of a TCP connection.
+/// \c IOEndpoint that represents an endpoint of a TCP packet.
 ///
-/// In the current implementation, an object of this class is always
-/// instantiated within the wrapper routines.  Applications are expected to
-/// get access to the object via the abstract base class, \c IOEndpoint.
-/// This design may be changed when we generalize the wrapper interface.
-///
-/// Note: this implementation is optimized for the case where this object
-/// is created from an ASIO endpoint object in a receiving code path
-/// by avoiding to make a copy of the base endpoint.  For TCP it may not be
-/// a big deal, but when we receive UDP packets at a high rate, the copy
-/// overhead might be significant.
+/// Other notes about \c TCPEndpoint applies to this class, too.
 class TCPEndpoint : public IOEndpoint {
 public:
     ///
-    /// \name Constructors and Destructor
+    /// \name Constructors and Destructor.
     ///
     //@{
+
+    /// \brief Default Constructor
+    ///
+    /// Creates an internal endpoint.  This is expected to be set by some
+    /// external call.
+    TCPEndpoint() :
+        asio_endpoint_placeholder_(new asio::ip::tcp::endpoint()),
+        asio_endpoint_(*asio_endpoint_placeholder_)
+    {}
+
     /// \brief Constructor from a pair of address and port.
     ///
     /// \param address The IP address of the endpoint.
     /// \param port The TCP port number of the endpoint.
     TCPEndpoint(const IOAddress& address, const unsigned short port) :
         asio_endpoint_placeholder_(
-            new asio::ip::tcp::endpoint(
-                asio::ip::address::from_string(address.toText()), port)),
+            new asio::ip::tcp::endpoint(asio::ip::address::from_string(address.toText()),
+                              port)),
         asio_endpoint_(*asio_endpoint_placeholder_)
     {}
 
@@ -59,39 +60,53 @@ public:
     /// corresponding ASIO class, \c tcp::endpoint.
     ///
     /// \param asio_endpoint The ASIO representation of the TCP endpoint.
-    TCPEndpoint(const asio::ip::tcp::endpoint& asio_endpoint) :
+    TCPEndpoint(asio::ip::tcp::endpoint& asio_endpoint) :
         asio_endpoint_placeholder_(NULL), asio_endpoint_(asio_endpoint)
     {}
 
+    /// \brief Constructor from an ASIO TCP endpoint.
+    ///
+    /// This constructor is designed to be an efficient wrapper for the
+    /// corresponding ASIO class, \c tcp::endpoint.
+    ///
+    /// \param asio_endpoint The ASIO representation of the TCP endpoint.
+    TCPEndpoint(const asio::ip::tcp::endpoint& asio_endpoint) :
+        asio_endpoint_placeholder_(new asio::ip::tcp::endpoint(asio_endpoint)),
+        asio_endpoint_(*asio_endpoint_placeholder_)
+    {}
+
     /// \brief The destructor.
-    ~TCPEndpoint() { delete asio_endpoint_placeholder_; }
+    virtual ~TCPEndpoint() { delete asio_endpoint_placeholder_; }
     //@}
 
-    IOAddress getAddress() const {
+    virtual IOAddress getAddress() const {
         return (asio_endpoint_.address());
     }
 
-    uint16_t getPort() const {
+    virtual uint16_t getPort() const {
         return (asio_endpoint_.port());
     }
 
-    short getProtocol() const {
+    virtual short getProtocol() const {
         return (asio_endpoint_.protocol().protocol());
     }
 
-    short getFamily() const {
+    virtual short getFamily() const {
         return (asio_endpoint_.protocol().family());
     }
 
     // This is not part of the exosed IOEndpoint API but allows
     // direct access to the ASIO implementation of the endpoint
-    const asio::ip::tcp::endpoint& getASIOEndpoint() const {
+    inline const asio::ip::tcp::endpoint& getASIOEndpoint() const {
+        return (asio_endpoint_);
+    }
+    inline asio::ip::tcp::endpoint& getASIOEndpoint() {
         return (asio_endpoint_);
     }
 
 private:
-    const asio::ip::tcp::endpoint* asio_endpoint_placeholder_;
-    const asio::ip::tcp::endpoint& asio_endpoint_;
+    asio::ip::tcp::endpoint* asio_endpoint_placeholder_;
+    asio::ip::tcp::endpoint& asio_endpoint_;
 };
 
 }      // namespace asiolink

src/lib/asiolink/tcp_server.cc

@@ -17,6 +17,7 @@
 #include <netinet/in.h>
 #include <sys/socket.h>
 #include <unistd.h>             // for some IPC/network system calls
+#include <errno.h>
 
 #include <boost/shared_array.hpp>
 
@@ -77,23 +78,21 @@ TCPServer::operator()(error_code ec, size_t length) {
             /// Create a socket to listen for connections
             socket_.reset(new tcp::socket(acceptor_->get_io_service()));
 
-            /// Wait for new connections. In the event of error,
+            /// Wait for new connections. In the event of non-fatal error,
             /// try again
             do {
                 CORO_YIELD acceptor_->async_accept(*socket_, *this);
 
-                // return if we meet fatal error
-                // Todo add log
+                // Abort on fatal errors
+                // TODO: Log error?
                 if (ec) {
                     using namespace asio::error;
                     if (ec.value() != would_block && ec.value() != try_again &&
-                            ec.value() != connection_aborted &&
-                            ec.value() != interrupted) {
+                        ec.value() != connection_aborted &&
+                        ec.value() != interrupted) {
                         return;
                     }
                 }
-
-
             } while (ec);
 
             /// Fork the coroutine by creating a copy of this one and

src/lib/asiolink/tcp_socket.h

@@ -24,11 +24,18 @@
 #include <sys/socket.h>
 #include <unistd.h>             // for some IPC/network system calls
 
-#include <iostream>
+#include <algorithm>
+#include <cassert>
 #include <cstddef>
 
+#include <boost/bind.hpp>
+#include <boost/numeric/conversion/cast.hpp>
+
 #include <config.h>
 
+#include <dns/buffer.h>
+
+#include <asiolink/asiolink_utilities.h>
 #include <asiolink/io_asio_socket.h>
 #include <asiolink/io_endpoint.h>
 #include <asiolink/io_service.h>
@@ -36,6 +43,15 @@
 
 namespace asiolink {
 
+/// \brief Buffer Too Large
+///
+/// Thrown on an attempt to send a buffer > 64k
+class BufferTooLarge : public IOError {
+public:
+    BufferTooLarge(const char* file, size_t line, const char* what) :
+        IOError(file, line, what) {}
+};
+
 /// \brief The \c TCPSocket class is a concrete derived class of \c IOAsioSocket
 /// that represents a TCP socket.
 ///
@@ -48,18 +64,18 @@ private:
     TCPSocket& operator=(const TCPSocket&);
 
 public:
-    
+
     /// \brief Constructor from an ASIO TCP socket.
     ///
-    /// \param socket The ASIO representation of the TCP socket.  It
-    /// is assumed that the caller will open and close the socket, so
-    /// these operations are a no-op for that socket.
+    /// \param socket The ASIO representation of the TCP socket.  It is assumed
+    ///        that the caller will open and close the socket, so these
+    ///        operations are a no-op for that socket.
     TCPSocket(asio::ip::tcp::socket& socket);
 
     /// \brief Constructor
     ///
     /// Used when the TCPSocket is being asked to manage its own internal
-    /// socket.  It is assumed that open() and close() will not be used.
+    /// socket.  In this case, the open() and close() methods are used.
     ///
     /// \param service I/O Service object used to manage the socket.
     TCPSocket(IOService& service);
@@ -67,68 +83,79 @@ public:
     /// \brief Destructor
     virtual ~TCPSocket();
 
-    virtual int getNative() const { return (socket_.native()); }
-    virtual int getProtocol() const { return (IPPROTO_TCP); }
+    /// \brief Return file descriptor of underlying socket
+    virtual int getNative() const {
+        return (socket_.native());
+    }
 
-    /// \brief Open Socket
+    /// \brief Return protocol of socket
+    virtual int getProtocol() const {
+        return (IPPROTO_TCP);
+    }
+
+    /// \brief Is "open()" synchronous?
     ///
-    /// Opens the TCP socket.  In the model for transport-layer agnostic I/O,
-    /// an "open" operation includes a connection to the remote end (which
-    /// may take time).  This does not happen for TCP, so the method returns
-    /// "false" to indicate that the operation completed synchronously.
+    /// Indicates that the opening of a TCP socket is asynchronous.
+    virtual bool isOpenSynchronous() const {
+        return (false);
+    }
+
+    /// \brief Open Socket
     ///
-    /// \param endpoint Endpoint to which the socket will connect to.
-    /// \param callback Unused.
+    /// Opens the TCP socket.  This is an asynchronous operation, completion of
+    /// which will be signalled via a call to the callback function.
     ///
-    /// \return false to indicate that the "operation" completed synchronously.
-    virtual bool open(const IOEndpoint* endpoint, C&);
+    /// \param endpoint Endpoint to which the socket will connect.
+    /// \param callback Callback object.
+    virtual void open(const IOEndpoint* endpoint, C& callback);
 
     /// \brief Send Asynchronously
     ///
-    /// This corresponds to async_send_to() for TCP sockets and async_send()
-    /// for TCP.  In both cases an endpoint argument is supplied indicating the
-    /// target of the send - this is ignored for TCP.
+    /// Calls the underlying socket's async_send() method to send a packet of
+    /// data asynchronously to the remote endpoint.  The callback will be called
+    /// on completion.
     ///
     /// \param data Data to send
     /// \param length Length of data to send
-    /// \param endpoint Target of the send
+    /// \param endpoint Target of the send. (Unused for a TCP socket because
+    ///        that was determined when the connection was opened.)
     /// \param callback Callback object.
     virtual void asyncSend(const void* data, size_t length,
-        const IOEndpoint* endpoint, C& callback);
+                           const IOEndpoint* endpoint, C& callback);
 
     /// \brief Receive Asynchronously
     ///
-    /// This correstponds to async_receive_from() for TCP sockets and
-    /// async_receive() for TCP.  In both cases, an endpoint argument is
-    /// supplied to receive the source of the communication.  For TCP it will
-    /// be filled in with details of the connection.
+    /// Calls the underlying socket's async_receive() method to read a packet
+    /// of data from a remote endpoint.  Arrival of the data is signalled via a
+    /// call to the callback function.
     ///
     /// \param data Buffer to receive incoming message
     /// \param length Length of the data buffer
-    /// \param cumulative Amount of data that should already be in the buffer.
-    /// (This is ignored - every UPD receive fills the buffer from the start.)
+    /// \param offset Offset into buffer where data is to be put
     /// \param endpoint Source of the communication
     /// \param callback Callback object
-    virtual void asyncReceive(void* data, size_t length, size_t cumulative,
-        IOEndpoint* endpoint, C& callback);
+    virtual void asyncReceive(void* data, size_t length, size_t offset,
+                              IOEndpoint* endpoint, C& callback);
 
-    /// \brief Checks if the data received is complete.
+    /// \brief Process received data packet
     ///
-    /// As all the data is received in one I/O, so this is, this is effectively
-    /// a no-op (although it does update the amount of data received).
+    /// See the description of IOAsioSocket::receiveComplete for a complete
+    /// description of this method.
     ///
-    /// \param data Data buffer containing data to date.  (This is ignored
-    /// for TCP receives.)
-    /// \param length Amount of data received in last asynchronous I/O
-    /// \param cumulative On input, amount of data received before the last
-    /// I/O.  On output, the total amount of data received to date.
+    /// \param staging Pointer to the start of the staging buffer.
+    /// \param length Amount of data in the staging buffer.
+    /// \param cumulative Amount of data received before the staging buffer is
+    ///        processed.
+    /// \param offset Unused.
+    /// \param expected unused.
+    /// \param outbuff Output buffer.  Data in the staging buffer is be copied
+    ///        to this output buffer in the call.
     ///
-    /// \return true if the receive is complete, false if another receive is
-    /// needed.
-    virtual bool receiveComplete(void*, size_t length, size_t& cumulative) {
-        cumulative = length;
-        return (true);
-    }
+    /// \return Always true
+    virtual bool processReceivedData(const void* staging, size_t length,
+                                     size_t& cumulative, size_t& offset,
+                                     size_t& expected,
+                                     isc::dns::OutputBufferPtr& outbuff);
 
     /// \brief Cancel I/O On Socket
     virtual void cancel();
@@ -144,13 +171,28 @@ private:
     asio::ip::tcp::socket*      socket_ptr_;    ///< Pointer to own socket
     asio::ip::tcp::socket&      socket_;        ///< Socket
     bool                        isopen_;        ///< true when socket is open
+
+    // TODO: Remove temporary buffer
+    // The current implementation copies the buffer passed to asyncSend() into
+    // a temporary buffer and precedes it with a two-byte count field.  As
+    // ASIO should really be just about sending and receiving data, the TCP
+    // code should not do this.  If the protocol using this requires a two-byte
+    // count, it should add it before calling this code.  (This may be best
+    // achieved by altering isc::dns::buffer to have pairs of methods:
+    // getLength()/getTCPLength(), getData()/getTCPData(), with the getTCPXxx()
+    // methods taking into account a two-byte count field.)
+    //
+    // The option of sending the data in two operations, the count followed by
+    // the data was discounted as that would lead to two callbacks which would
+    // cause problems with the stackless coroutine code.
+    isc::dns::OutputBufferPtr   send_buffer_;   ///< Send buffer
 };
 
 // Constructor - caller manages socket
 
 template <typename C>
 TCPSocket<C>::TCPSocket(asio::ip::tcp::socket& socket) :
-    socket_ptr_(NULL), socket_(socket), isopen_(true)
+    socket_ptr_(NULL), socket_(socket), isopen_(true), send_buffer_()
 {
 }
 
@@ -171,16 +213,14 @@ TCPSocket<C>::~TCPSocket()
     delete socket_ptr_;
 }
 
-// Open the socket.  Throws an error on failure
-// TODO: Make the open more resilient
+// Open the socket.
 
-template <typename C> bool
-TCPSocket<C>::open(const IOEndpoint* endpoint, C&) {
+template <typename C> void
+TCPSocket<C>::open(const IOEndpoint* endpoint, C& callback) {
 
     // Ignore opens on already-open socket.  Don't throw a failure because
     // of uncertainties as to what precedes whan when using asynchronous I/O.
     // At also allows us a treat a passed-in socket as a self-managed socket.
-
     if (!isopen_) {
         if (endpoint->getFamily() == AF_INET) {
             socket_.open(asio::ip::tcp::v4());
@@ -190,10 +230,25 @@ TCPSocket<C>::open(const IOEndpoint* endpoint, C&) {
         }
         isopen_ = true;
 
-        // TODO: Complete TCPSocket::open()
+        // Set options on the socket:
 
+        // Reuse address - allow the socket to bind to a port even if the port
+        // is in the TIMED_WAIT state.
+        socket_.set_option(asio::socket_base::reuse_address(true));
     }
-    return (false);
+
+    // Upconvert to a TCPEndpoint.  We need to do this because although
+    // IOEndpoint is the base class of UDPEndpoint and TCPEndpoint, it does not
+    // contain a method for getting at the underlying endpoint type - that is in
+    /// the derived class and the two classes differ on return type.
+    assert(endpoint->getProtocol() == IPPROTO_TCP);
+    const TCPEndpoint* tcp_endpoint =
+        static_cast<const TCPEndpoint*>(endpoint);
+
+    // Connect to the remote endpoint.  On success, the handler will be
+    // called (with one argument - the length argument will default to
+    // zero).
+    socket_.async_connect(tcp_endpoint->getASIOEndpoint(), callback);
 }
 
 // Send a message.  Should never do this if the socket is not open, so throw
@@ -201,24 +256,29 @@ TCPSocket<C>::open(const IOEndpoint* endpoint, C&) {
 
 template <typename C> void
 TCPSocket<C>::asyncSend(const void* data, size_t length,
-    const IOEndpoint* endpoint, C& callback)
+    const IOEndpoint*, C& callback)
 {
     if (isopen_) {
 
-        // Upconvert to a TCPEndpoint.  We need to do this because although
-        // IOEndpoint is the base class of TCPEndpoint and TCPEndpoint, it
-        // doing cont contain a method for getting at the underlying endpoint
-        // type - those are in the derived class and the two classes differ on
-        // return type.
-
-        assert(endpoint->getProtocol() == IPPROTO_TCP);
-        const TCPEndpoint* tcp_endpoint =
-            static_cast<const TCPEndpoint*>(endpoint);
-        std::cerr << "TCPSocket::asyncSend(): sending to " <<
-            tcp_endpoint->getAddress().toText() <<
-            ", port " << tcp_endpoint->getPort() << "\n";
-
-        // TODO: Complete TCPSocket::asyncSend()
+        // Need to copy the data into a temporary buffer and precede it with
+        // a two-byte count field.
+        // TODO: arrange for the buffer passed to be preceded by the count
+        try {
+            // Ensure it fits into 16 bits
+            uint16_t count = boost::numeric_cast<uint16_t>(length);
+
+            // Copy data into a buffer preceded by the count field.
+            send_buffer_.reset(new isc::dns::OutputBuffer(length + 2));
+            send_buffer_->writeUint16(count);
+            send_buffer_->writeData(data, length);
+
+            // ... and send it
+            socket_.async_send(asio::buffer(send_buffer_->getData(),
+                               send_buffer_->getLength()), callback);
+        } catch (boost::numeric::bad_numeric_cast& e) {
+            isc_throw(BufferTooLarge,
+                      "attempt to send buffer larger than 64kB");
+        }
 
     } else {
         isc_throw(SocketNotOpen,
@@ -226,26 +286,40 @@ TCPSocket<C>::asyncSend(const void* data, size_t length,
     }
 }
 
-// Receive a message. Note that the "cumulative" argument is ignored - every TCP
-// receive is put into the buffer beginning at the start - there is no concept
-// receiving a subsequent part of a message.  Same critera as before concerning
-// the need for the socket to be open.
-
+// Receive a message. Note that the "offset" argument is used as an index
+// into the buffer in order to decide where to put the data.  It is up to the
+// caller to initialize the data to zero
 template <typename C> void
-TCPSocket<C>::asyncReceive(void* data, size_t length, size_t,
+TCPSocket<C>::asyncReceive(void* data, size_t length, size_t offset,
     IOEndpoint* endpoint, C& callback)
 {
     if (isopen_) {
-
-        // Upconvert the endpoint again.
+        // Upconvert to a TCPEndpoint.  We need to do this because although
+        // IOEndpoint is the base class of UDPEndpoint and TCPEndpoint, it
+        // does not contain a method for getting at the underlying endpoint
+        // type - that is in the derived class and the two classes differ on
+        // return type.
         assert(endpoint->getProtocol() == IPPROTO_TCP);
-        const TCPEndpoint* tcp_endpoint =
-            static_cast<const TCPEndpoint*>(endpoint);
-        std::cerr << "TCPSocket::asyncReceive(): receiving from " <<
-            tcp_endpoint->getAddress().toText() <<
-            ", port " << tcp_endpoint->getPort() << "\n";
+        TCPEndpoint* tcp_endpoint = static_cast<TCPEndpoint*>(endpoint);
+
+        // Write the endpoint details from the communications link.  Ideally
+        // we should make IOEndpoint assignable, but this runs in to all sorts
+        // of problems concerning the management of the underlying Boost
+        // endpoint (e.g. if it is not self-managed, is the copied one
+        // self-managed?) The most pragmatic solution is to let Boost take care
+        // of everything and copy details of the underlying endpoint.
+        tcp_endpoint->getASIOEndpoint() = socket_.remote_endpoint();
+
+        // Ensure we can write into the buffer and if so, set the pointer to
+        // where the data will be written.
+        if (offset >= length) {
+            isc_throw(BufferOverflow, "attempt to read into area beyond end of "
+                                      "TCP receive buffer");
+        }
+        void* buffer_start = static_cast<void*>(static_cast<uint8_t*>(data) + offset);
 
-        // TODO: Complete TCPSocket::asyncReceive()
+        // ... and kick off the read.
+        socket_.async_receive(asio::buffer(buffer_start, length - offset), callback);
 
     } else {
         isc_throw(SocketNotOpen,
@@ -253,7 +327,72 @@ TCPSocket<C>::asyncReceive(void* data, size_t length, size_t,
     }
 }
 
+// Is the receive complete?
+
+template <typename C> bool
+TCPSocket<C>::processReceivedData(const void* staging, size_t length,
+                                  size_t& cumulative, size_t& offset,
+                                  size_t& expected,
+                                  isc::dns::OutputBufferPtr& outbuff)
+{
+    // Point to the data in the staging buffer and note how much there is.
+    const uint8_t* data = static_cast<const uint8_t*>(staging);
+    size_t data_length = length;
+
+    // Is the number is "expected" valid?  It won't be unless we have received
+    // at least two bytes of data in total for this set of receives.
+    if (cumulative < 2) {
+
+        // "expected" is not valid.  Did this read give us enough data to
+        // work it out?
+        cumulative += length;
+        if (cumulative < 2) {
+
+            // Nope, still not valid.  This must have been the first packet and
+            // was only one byte long.  Tell the fetch code to read the next
+            // packet into the staging buffer beyond the data that is already
+            // there so that the next time we are called we have a complete
+            // TCP count.
+            offset = cumulative;
+            return (false);
+        }
+
+        // Have enough data to interpret the packet count, so do so now.
+        expected = readUint16(data);
+
+        // We have two bytes less of data to process.  Point to the start of the
+        // data and adjust the packet size.  Note that at this point,
+        // "cumulative" is the true amount of data in the staging buffer, not
+        // "length".
+        data += 2;
+        data_length = cumulative - 2;
+    } else {
+
+        // Update total amount of data received.
+        cumulative += length;
+    }
+
+    // Regardless of anything else, the next read goes into the start of the
+    // staging buffer.
+    offset = 0;
+
+    // Work out how much data we still have to put in the output buffer. (This
+    // could be zero if we have just interpreted the TCP count and that was
+    // set to zero.)
+    if (expected >= outbuff->getLength()) {
+
+        // Still need data in the output packet.  Copy what we can from the
+        // staging buffer to the output buffer.
+        size_t copy_amount = std::min(expected - outbuff->getLength(), data_length);
+        outbuff->writeData(data, copy_amount);
+    }
+
+    // We can now say if we have all the data.
+    return (expected == outbuff->getLength());
+}
+
 // Cancel I/O on the socket.  No-op if the socket is not open.
+
 template <typename C> void
 TCPSocket<C>::cancel() {
     if (isopen_) {

src/lib/asiolink/tests/Makefile.am

@@ -18,13 +18,15 @@ TESTS += run_unittests
 run_unittests_SOURCES  = run_unittests.cc
 run_unittests_SOURCES += $(top_srcdir)/src/lib/dns/tests/unittest_util.h
 run_unittests_SOURCES += $(top_srcdir)/src/lib/dns/tests/unittest_util.cc
+run_unittests_SOURCES += asiolink_utilities_unittest.cc
 run_unittests_SOURCES += io_address_unittest.cc
 run_unittests_SOURCES += io_endpoint_unittest.cc
 run_unittests_SOURCES += io_fetch_unittest.cc
 run_unittests_SOURCES += io_socket_unittest.cc
 run_unittests_SOURCES += io_service_unittest.cc
 run_unittests_SOURCES += interval_timer_unittest.cc
-run_unittests_SOURCES += recursive_query_unittest.cc
+run_unittests_SOURCES += tcp_endpoint_unittest.cc
+run_unittests_SOURCES += tcp_socket_unittest.cc
 run_unittests_SOURCES += udp_endpoint_unittest.cc
 run_unittests_SOURCES += udp_socket_unittest.cc
 run_unittests_SOURCES += dns_server_unittest.cc
@@ -34,12 +36,10 @@ run_unittests_CPPFLAGS = $(AM_CPPFLAGS) $(GTEST_INCLUDES)
 
 run_unittests_LDADD  = $(GTEST_LDADD)
 run_unittests_LDADD += $(SQLITE_LIBS)
-run_unittests_LDADD +=  $(top_builddir)/src/lib/dns/libdns++.la
-run_unittests_LDADD += $(top_builddir)/src/lib/exceptions/libexceptions.la
+run_unittests_LDADD += $(top_builddir)/src/lib/dns/libdns++.la
 run_unittests_LDADD += $(top_builddir)/src/lib/asiolink/libasiolink.la
 run_unittests_LDADD += $(top_builddir)/src/lib/log/liblog.la
-run_unittests_LDADD += $(top_builddir)/src/lib/cache/libcache.la
-run_unittests_LDADD += $(top_builddir)/src/lib/nsas/libnsas.la
+run_unittests_LDADD += $(top_builddir)/src/lib/exceptions/libexceptions.la
 
 run_unittests_LDFLAGS = $(AM_LDFLAGS) $(GTEST_LDFLAGS) 
 

src/lib/asiolink/tests/asiolink_utilities_unittest.cc

@@ -0,0 +1,74 @@
+// Copyright (C) 2011  Internet Systems Consortium, Inc. ("ISC")
+//
+// Permission to use, copy, modify, and/or distribute this software for any
+// purpose with or without fee is hereby granted, provided that the above
+// copyright notice and this permission notice appear in all copies.
+//
+// THE SOFTWARE IS PROVIDED "AS IS" AND ISC DISCLAIMS ALL WARRANTIES WITH
+// REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
+// AND FITNESS.  IN NO EVENT SHALL ISC BE LIABLE FOR ANY SPECIAL, DIRECT,
+// INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
+// LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
+// OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
+// PERFORMANCE OF THIS SOFTWARE.
+
+/// \brief Test of asiolink utilties
+///
+/// Tests the fuctionality of the asiolink utilities code by comparing them
+/// with the equivalent methods in isc::dns::[Input/Output]Buffer.
+
+#include <cstddef>
+
+#include <gtest/gtest.h>
+
+#include <dns/buffer.h>
+#include <asiolink/asiolink_utilities.h>
+
+using namespace asiolink;
+using namespace isc::dns;
+
+TEST(asioutil, readUint16) {
+
+    // Reference buffer
+    uint8_t data[2];
+    isc::dns::InputBuffer buffer(data, sizeof(data));
+
+    // Avoid possible compiler warnings by only setting uint8_t variables to
+    // uint8_t values.
+    uint8_t i8 = 0;
+    uint8_t j8 = 0;
+    for (int i = 0; i < (2 << 8); ++i, ++i8) {
+        for (int j = 0; j < (2 << 8); ++j, ++j8) {
+            data[0] = i8;
+            data[1] = j8;
+            buffer.setPosition(0);
+            EXPECT_EQ(buffer.readUint16(), readUint16(data));
+        }
+    }
+}
+
+
+TEST(asioutil, writeUint16) {
+
+    // Reference buffer
+    isc::dns::OutputBuffer buffer(2);
+    uint8_t test[2];
+
+    // Avoid possible compiler warnings by only setting uint16_t variables to
+    // uint16_t values.
+    uint16_t i16 = 0;
+    for (uint32_t i = 0; i < (2 << 16); ++i, ++i16) {
+
+        // Write the reference data
+        buffer.clear();
+        buffer.writeUint16(i16);
+
+        // ... and the test data
+        writeUint16(i16, test);
+
+        // ... and compare
+        const uint8_t* ref = static_cast<const uint8_t*>(buffer.getData());
+        EXPECT_EQ(ref[0], test[0]);
+        EXPECT_EQ(ref[1], test[1]);
+    }
+}

src/lib/asiolink/tests/io_fetch_unittest.cc

@@ -12,13 +12,17 @@
 // OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
 // PERFORMANCE OF THIS SOFTWARE.
 
-#include <gtest/gtest.h>
-#include <boost/bind.hpp>
+#include <algorithm>
 #include <cstdlib>
 #include <string>
+#include <iostream>
+#include <iomanip>
+#include <iterator>
 #include <vector>
 
-#include <string.h>
+#include <gtest/gtest.h>
+#include <boost/bind.hpp>
+#include <boost/date_time/posix_time/posix_time_types.hpp>
 
 #include <asio.hpp>
 
@@ -30,19 +34,27 @@
 #include <dns/name.h>
 #include <dns/rcode.h>
 
+#include <asiolink/asiolink_utilities.h>
+#include <asiolink/io_address.h>
+#include <asiolink/io_endpoint.h>
 #include <asiolink/io_fetch.h>
 #include <asiolink/io_service.h>
 
 using namespace asio;
 using namespace isc::dns;
-using asio::ip::udp;
+using namespace asio::ip;
+using namespace std;
 
 namespace asiolink {
 
 const asio::ip::address TEST_HOST(asio::ip::address::from_string("127.0.0.1"));
 const uint16_t TEST_PORT(5301);
-// FIXME Shouldn't we send something that is real message?
-const char TEST_DATA[] = "TEST DATA";
+const int SEND_INTERVAL = 250;      // Interval in ms between TCP sends
+const size_t MAX_SIZE = 64 * 1024;  // Should be able to take 64kB
+
+// The tests are complex, so debug output has been left in (although disabled).
+// Set this to true to enable it.
+const bool DEBUG = false;
 
 /// \brief Test fixture for the asiolink::IOFetch.
 class IOFetchTest : public virtual ::testing::Test, public virtual IOFetch::Callback
@@ -52,13 +64,26 @@ public:
     IOFetch::Result expected_;      ///< Expected result of the callback
     bool            run_;           ///< Did the callback run already?
     Question        question_;      ///< What to ask
-    OutputBufferPtr buff_;          ///< Buffer to hold result
+    OutputBufferPtr result_buff_;   ///< Buffer to hold result of fetch
+    OutputBufferPtr msgbuf_;        ///< Buffer corresponding to known question
     IOFetch         udp_fetch_;     ///< For UDP query test
-    //IOFetch         tcp_fetch_;     ///< For TCP query test
-
-    // The next member is the buffer iin which the "server" (implemented by the
-    // response handler method) receives the question sent by the fetch object.
-    std::vector<char>      server_buff_;  ///< Server buffer
+    IOFetch         tcp_fetch_;     ///< For TCP query test
+    IOFetch::Protocol protocol_;    ///< Protocol being tested
+    size_t          cumulative_;    ///< Cumulative data received by "server".
+    deadline_timer  timer_;         ///< Timer to measure timeouts
+
+    // The next member is the buffer in which the "server" (implemented by the
+    // response handler methods in this class) receives the question sent by the
+    // fetch object.
+    uint8_t         receive_buffer_[MAX_SIZE]; ///< Server receive buffer
+    vector<uint8_t> send_buffer_;           ///< Server send buffer
+    uint16_t        send_cumulative_;       ///< Data sent so far
+
+    // Other data.
+    string          return_data_;           ///< Data returned by server
+    string          test_data_;             ///< Large string - here for convenience
+    bool            debug_;                 ///< true to enable debug output
+    size_t          tcp_send_size_;         ///< Max size of TCP send
 
     /// \brief Constructor
     IOFetchTest() :
@@ -66,126 +91,518 @@ public:
         expected_(IOFetch::NOTSET),
         run_(false),
         question_(Name("example.net"), RRClass::IN(), RRType::A()),
-        buff_(new OutputBuffer(512)),
-        udp_fetch_(IPPROTO_UDP, service_, question_, IOAddress(TEST_HOST),
-                   TEST_PORT, buff_, this, 100),
-        server_buff_(512)
-        // tcp_fetch_(service_, question_, IOAddress(TEST_HOST), TEST_PORT,
-        //    buff_, this, 100, IPPROTO_UDP)
-        { }
+        result_buff_(new OutputBuffer(512)),
+        msgbuf_(new OutputBuffer(512)),
+        udp_fetch_(IOFetch::UDP, service_, question_, IOAddress(TEST_HOST),
+            TEST_PORT, result_buff_, this, 100),
+        tcp_fetch_(IOFetch::TCP, service_, question_, IOAddress(TEST_HOST),
+            TEST_PORT, result_buff_, this, (16 * SEND_INTERVAL)),
+                                        // Timeout interval chosen to ensure no timeout
+        protocol_(IOFetch::TCP),        // for initialization - will be changed
+        cumulative_(0),
+        timer_(service_.get_io_service()),
+        receive_buffer_(),
+        send_buffer_(),
+        send_cumulative_(0),
+        return_data_(""),
+        test_data_(""),
+        debug_(DEBUG),
+        tcp_send_size_(0)
+    {
+        // Construct the data buffer for question we expect to receive.
+        Message msg(Message::RENDER);
+        msg.setQid(0);
+        msg.setOpcode(Opcode::QUERY());
+        msg.setRcode(Rcode::NOERROR());
+        msg.setHeaderFlag(Message::HEADERFLAG_RD);
+        msg.addQuestion(question_);
+        MessageRenderer renderer(*msgbuf_);
+        msg.toWire(renderer);
+
+        // Initialize the test data to be returned: tests will return a
+        // substring of this data. (It's convenient to have this as a member of
+        // the class.)
+        //
+        // We could initialize the data with a single character, but as an added
+        // check we'll make ssre that it has some structure.
+
+        test_data_.clear();
+        test_data_.reserve(MAX_SIZE);
+        while (test_data_.size() < MAX_SIZE) {
+            test_data_ += "A message to be returned to the client that has "
+                          "some sort of structure.";
+        }
+    }
+
+    /// \brief UDP Response handler (the "remote UDP DNS server")
+    ///
+    /// When IOFetch is sending data, this response handler emulates the remote
+    /// DNS server.  It checks that the data sent by the IOFetch object is what
+    /// was expected to have been sent, then sends back a known buffer of data.
+    ///
+    /// \param remote Endpoint to which to send the answer
+    /// \param socket Socket to use to send the answer
+    /// \param ec ASIO error code, completion code of asynchronous I/O issued
+    ///        by the "server" to receive data.
+    /// \param length Amount of data received.
+    void udpReceiveHandler(udp::endpoint* remote, udp::socket* socket,
+                    error_code ec = error_code(), size_t length = 0) {
+        if (debug_) {
+            cout << "udpReceiveHandler(): error = " << ec.value() <<
+                    ", length = " << length << endl;
+        }
+
+        // The QID in the incoming data is random so set it to 0 for the
+        // data comparison check. (It is set to 0 in the buffer containing
+        // the expected data.)
+        receive_buffer_[0] = receive_buffer_[1] = 0;
+
+        // Check that length of the received data and the expected data are
+        // identical, then check that the data is identical as well.
+        EXPECT_EQ(msgbuf_->getLength(), length);
+        EXPECT_TRUE(equal(receive_buffer_, (receive_buffer_ + length - 1),
+        static_cast<const uint8_t*>(msgbuf_->getData())));
+
+        // Return a message back to the IOFetch object.
+        socket->send_to(asio::buffer(return_data_.c_str(), return_data_.size()),
+                                     *remote);
+        if (debug_) {
+            cout << "udpReceiveHandler(): returned " << return_data_.size() <<
+                    " bytes to the client" << endl;
+        }
+    }
+
+    /// \brief Completion Handler for accepting TCP data
+    ///
+    /// Called when the remote system connects to the "server".  It issues
+    /// an asynchronous read on the socket to read data.
+    ///
+    /// \param socket Socket on which data will be received
+    /// \param ec Boost error code, value should be zero.
+    void tcpAcceptHandler(tcp::socket* socket, error_code ec = error_code())
+    {
+        if (debug_) {
+            cout << "tcpAcceptHandler(): error = " << ec.value() << endl;
+        }
+
+        // Expect that the accept completed without a problem.
+        EXPECT_EQ(0, ec.value());
+
+        // Work out the maximum size of data we can send over it when we
+        // respond, then subtract 1kB or so for safety.
+        tcp::socket::send_buffer_size send_size;
+        socket->get_option(send_size);
+        if (send_size.value() < (2 * 1024)) {
+            FAIL() << "TCP send size is less than 2kB";
+        } else {
+            tcp_send_size_ = send_size.value() - 1024;
+            if (debug_) {
+                cout << "tcpacceptHandler(): will use send size = " << tcp_send_size_ << endl;
+            }
+        }
+
+        // Initiate a read on the socket.
+        cumulative_ = 0;
+        socket->async_receive(asio::buffer(receive_buffer_, sizeof(receive_buffer_)),
+            boost::bind(&IOFetchTest::tcpReceiveHandler, this, socket, _1, _2));
+    }
+
+    /// \brief Completion handler for receiving TCP data
+    ///
+    /// When IOFetch is sending data, this response handler emulates the remote
+    /// DNS server.  It that all the data sent by the IOFetch object has been
+    /// received, issuing another read if not.  If the data is complete, it is
+    /// compared to what is expected and a reply sent back to the IOFetch.
+    ///
+    /// \param socket Socket to use to send the answer
+    /// \param ec ASIO error code, completion code of asynchronous I/O issued
+    ///        by the "server" to receive data.
+    /// \param length Amount of data received.
+    void tcpReceiveHandler(tcp::socket* socket, error_code ec = error_code(),
+                           size_t length = 0)
+    {
+        if (debug_) {
+            cout << "tcpReceiveHandler(): error = " << ec.value() <<
+                    ", length = " << length << endl;
+        }
+        // Expect that the receive completed without a problem.
+        EXPECT_EQ(0, ec.value());
+
+        // If we haven't received all the data, issue another read.
+        cumulative_ += length;
+        bool complete = false;
+        if (cumulative_ > 2) {
+            uint16_t dns_length = readUint16(receive_buffer_);
+            complete = ((dns_length + 2) == cumulative_);
+        }
+
+        if (!complete) {
+            socket->async_receive(asio::buffer((receive_buffer_ + cumulative_),
+                (sizeof(receive_buffer_) - cumulative_)),
+                boost::bind(&IOFetchTest::tcpReceiveHandler, this, socket, _1, _2));
+            return;
+        }
+
+        // Check that length of the DNS message received is that expected, then
+        // compare buffers, zeroing the QID in the received buffer to match
+        // that set in our expected question.  Note that due to the length
+        // field the QID in the received buffer is in the third and fourth
+        // bytes.
+        EXPECT_EQ(msgbuf_->getLength() + 2, cumulative_);
+        receive_buffer_[2] = receive_buffer_[3] = 0;
+        EXPECT_TRUE(equal((receive_buffer_ + 2), (receive_buffer_ + cumulative_ - 2),
+            static_cast<const uint8_t*>(msgbuf_->getData())));
+
+        // ... and return a message back.  This has to be preceded by a two-byte
+        // count field.
+        send_buffer_.clear();
+        send_buffer_.push_back(0);
+        send_buffer_.push_back(0);
+        writeUint16(return_data_.size(), &send_buffer_[0]);
+        copy(return_data_.begin(), return_data_.end(), back_inserter(send_buffer_));
+
+        // Send the data.  This is done in multiple writes with a delay between
+        // each to check that the reassembly of TCP packets from fragments works.
+        send_cumulative_ = 0;
+        tcpSendData(socket);
+    }
+
+    /// \brief Sent Data Over TCP
+    ///
+    /// Send the TCP data back to the IOFetch object.  The data is sent in
+    /// three chunks - two of 16 bytes and the remainder, with a 250ms gap
+    /// between each. (Amounts of data smaller than one 32 bytes are sent in
+    /// one or two packets.)
+    ///
+    /// \param socket Socket over which send should take place
+    void tcpSendData(tcp::socket* socket) {
+        if (debug_) {
+            cout << "tcpSendData()" << endl;
+        }
+
+        // Decide what to send based on the cumulative count.  At most we'll do
+        // two chunks of 16 bytes (with a 250ms gap between) and then the
+        // remainder.
+        uint8_t* send_ptr = &send_buffer_[send_cumulative_];
+                                    // Pointer to data to send
+        size_t amount = 16;         // Amount of data to send
+        if (send_cumulative_ < (2 * amount)) {
+            
+            // First or second time through, send at most 16 bytes
+            amount = min(amount, (send_buffer_.size() - send_cumulative_));
+
+        } else {
+
+            // For all subsequent times, send the remainder, maximised to
+            // whatever we have chosen for the maximum send size.
+            amount = min(tcp_send_size_,
+                        (send_buffer_.size() - send_cumulative_));
+        }
+        if (debug_) {
+            cout << "tcpSendData(): sending " << amount << " bytes" << endl;
+        }
+
+
+        // ... and send it.  The amount sent is also passed as the first
+        // argument of the send callback, as a check.
+        socket->async_send(asio::buffer(send_ptr, amount),
+                           boost::bind(&IOFetchTest::tcpSendHandler, this,
+                                       amount, socket, _1, _2));
+    }
+
+    /// \brief Completion Handler for Sending TCP data
+    ///
+    /// Called when the asynchronous send of data back to the IOFetch object
+    /// by the TCP "server" in this class has completed.  (This send has to
+    /// be asynchronous because control needs to return to the caller in order
+    /// for the IOService "run()" method to be called to run the handlers.)
+    ///
+    /// If not all the data has been sent, a short delay is instigated (during
+    /// which control returns to the IOService).  This should force the queued
+    /// data to actually be sent and the IOFetch receive handler to be triggered.
+    /// In this way, the ability of IOFetch to handle fragmented TCP packets
+    /// should be checked.
+    ///
+    /// \param expected Number of bytes that were expected to have been sent.
+    /// \param socket Socket over which the send took place.  Only used to
+    ///        pass back to the send method.
+    /// \param ec Boost error code, value should be zero.
+    /// \param length Number of bytes sent.
+    void tcpSendHandler(size_t expected, tcp::socket* socket,
+                        error_code ec = error_code(), size_t length = 0)
+    {
+        if (debug_) {
+            cout << "tcpSendHandler(): error = " << ec.value() <<
+                    ", length = " << length << endl;
+        }
+
+        EXPECT_EQ(0, ec.value());       // Expect no error
+        EXPECT_EQ(expected, length);    // And that amount sent is as expected
+
+        // Do we need to send more?
+        send_cumulative_ += length;
+        if (send_cumulative_ < send_buffer_.size()) {
+
+            // Yes - set up a timer:  the callback handler for the timer is
+            // tcpSendData, which will then send the next chunk.  We pass the
+            // socket over which data should be sent as an argument to that
+            // function.
+            timer_.expires_from_now(boost::posix_time::milliseconds(SEND_INTERVAL));
+            timer_.async_wait(boost::bind(&IOFetchTest::tcpSendData, this,
+                                          socket));
+        }
+    }
 
     /// \brief Fetch completion callback
     ///
     /// This is the callback's operator() method which is called when the fetch
-    /// is complete.  Check that the data received is the wire format of the
-    /// question, then send back an arbitrary response.
+    /// is complete.  It checks that the data received is the wire format of the
+    /// data sent back by the server.
+    ///
+    /// \param result Result indicated by the callback
     void operator()(IOFetch::Result result) {
+        if (debug_) {
+            cout << "operator()(): result = " << result << endl;
+        }
+
         EXPECT_EQ(expected_, result);   // Check correct result returned
         EXPECT_FALSE(run_);             // Check it is run only once
         run_ = true;                    // Note success
-        service_.stop();                // ... and exit run loop
-    }
 
-    /// \brief Response handler, pretending to be remote DNS server
-    ///
-    /// This checks that the data sent is what we expected to receive, and
-    /// sends back a test answer.
-    void respond(udp::endpoint* remote, udp::socket* socket,
-            asio::error_code ec = asio::error_code(), size_t length = 0) {
+        // If the expected result for SUCCESS, then this should have been called
+        // when one of the "servers" in this class has sent back return_data_.
+        // Check the data is as expected/
+        if (expected_ == IOFetch::SUCCESS) {
+            EXPECT_EQ(return_data_.size(), result_buff_->getLength());
 
-        // Construct the data buffer for question we expect to receive.
-        OutputBuffer msgbuf(512);
-        Message msg(Message::RENDER);
-        msg.setQid(0);
-        msg.setOpcode(Opcode::QUERY());
-        msg.setRcode(Rcode::NOERROR());
-        msg.setHeaderFlag(Message::HEADERFLAG_RD);
-        msg.addQuestion(question_);
-        MessageRenderer renderer(msgbuf);
-        msg.toWire(renderer);
+            const uint8_t* start = static_cast<const uint8_t*>(result_buff_->getData());
+            EXPECT_TRUE(equal(return_data_.begin(), return_data_.end(), start));
+        }
 
-        // The QID in the incoming data is random so set it to 0 for the
-        // data comparison check. (It was set to 0 when the buffer containing
-        // the expected data was constructed above.)
-        server_buff_[0] = 0;
-        server_buff_[1] = 0;
+        // ... and cause the run loop to exit.
+        service_.stop();
+    }
 
-        // Check that lengths are identical.
-        EXPECT_EQ(msgbuf.getLength(), length);
-        EXPECT_TRUE(memcmp(msgbuf.getData(), &server_buff_[0], length) == 0);
+    // The next set of methods are the tests themselves.  A number of the TCP
+    // and UDP tests are very similar.
 
-        // ... and return a message back.
-        socket->send_to(asio::buffer(TEST_DATA, sizeof TEST_DATA), *remote);
+    /// \brief Check for stop()
+    ///
+    /// Test that when we run the query and stop it after it was run, it returns
+    /// "stopped" correctly. (That is why stop() is posted to the service_ as
+    /// well instead of calling it.)
+    ///
+    /// \param protocol Test protocol
+    /// \param fetch Fetch object being tested
+    void stopTest(IOFetch::Protocol protocol, IOFetch& fetch) {
+        protocol_ = protocol;
+        expected_ = IOFetch::STOPPED;
+
+        // Post the query
+        service_.get_io_service().post(fetch);
+
+        // Post query_.stop() (yes, the boost::bind thing is just
+        // query_.stop()).
+        service_.get_io_service().post(
+            boost::bind(&IOFetch::stop, fetch, IOFetch::STOPPED));
+
+        // Run both of them.  run() returns when everything in the I/O service
+        // queue has completed.
+        service_.run();
+        EXPECT_TRUE(run_);
     }
-};
 
+    /// \brief Premature stop test
+    ///
+    /// Test that when we queue the query to service_ and call stop() before it
+    /// gets executed, it acts sanely as well (eg. has the same result as
+    /// running stop() after - calls the callback).
+    ///
+    /// \param protocol Test protocol
+    /// \param fetch Fetch object being tested
+    void prematureStopTest(IOFetch::Protocol protocol, IOFetch& fetch) {
+        protocol_ = protocol;
+        expected_ = IOFetch::STOPPED;
+
+        // Stop before it is started
+        fetch.stop();
+        service_.get_io_service().post(fetch);
+
+        service_.run();
+        EXPECT_TRUE(run_);
+    }
 
-/// Test that when we run the query and stop it after it was run,
-/// it returns "stopped" correctly.
-///
-/// That is why stop() is posted to the service_ as well instead
-/// of calling it.
-TEST_F(IOFetchTest, UdpStop) {
-    expected_ = IOFetch::STOPPED;
+    /// \brief Timeout test
+    ///
+    /// Test that fetch times out when no answer arrives.
+    ///
+    /// \param protocol Test protocol
+    /// \param fetch Fetch object being tested
+    void timeoutTest(IOFetch::Protocol protocol, IOFetch& fetch) {
+        protocol_ = protocol;
+        expected_ = IOFetch::TIME_OUT;
+
+        service_.get_io_service().post(fetch);
+        service_.run();
+        EXPECT_TRUE(run_);
+    }
 
-    // Post the query
-    service_.get_io_service().post(udp_fetch_);
+    /// \brief Send/Receive Test
+    ///
+    /// Send a query to the server then receives a response.
+    ///
+    /// \param Test data to return to client
+    void tcpSendReturnTest(const std::string& return_data) {
+        if (debug_) {
+            cout << "tcpSendReturnTest(): data size = " << return_data.size() << endl;
+        }
+        return_data_ = return_data;
+        protocol_ = IOFetch::TCP;
+        expected_ = IOFetch::SUCCESS;
+
+        // Socket into which the connection will be accepted.
+        tcp::socket socket(service_.get_io_service());
+
+        // Acceptor object - called when the connection is made, the handler
+        // will initiate a read on the socket.
+        tcp::acceptor acceptor(service_.get_io_service(),
+                               tcp::endpoint(tcp::v4(), TEST_PORT));
+        acceptor.async_accept(socket,
+            boost::bind(&IOFetchTest::tcpAcceptHandler, this, &socket, _1));
+
+        // Post the TCP fetch object to send the query and receive the response.
+        service_.get_io_service().post(tcp_fetch_);
+
+        // ... and execute all the callbacks.  This exits when the fetch
+        // completes.
+        service_.run();
+        EXPECT_TRUE(run_);  // Make sure the callback did execute
+
+        // Tidy up
+        socket.close();
+    }
+};
 
-    // Post query_.stop() (yes, the boost::bind thing is just
-    // query_.stop()).
-    service_.get_io_service().post(
-       boost::bind(&IOFetch::stop, udp_fetch_, IOFetch::STOPPED));
+// Check the protocol
+TEST_F(IOFetchTest, Protocol) {
+    EXPECT_EQ(IOFetch::UDP, udp_fetch_.getProtocol());
+    EXPECT_EQ(IOFetch::TCP, tcp_fetch_.getProtocol());
+}
 
-    // Run both of them.  run() returns when everything in the I/O service
-    // queue has completed.
-    service_.run();
-    EXPECT_TRUE(run_);
+// UDP Stop test - see IOFetchTest::stopTest() header.
+TEST_F(IOFetchTest, UdpStop) {
+    stopTest(IOFetch::UDP, udp_fetch_);
 }
 
-// Test that when we queue the query to service_ and call stop() before it gets
-// executed, it acts sanely as well (eg. has the same result as running stop()
-// after - calls the callback).
+// UDP premature stop test - see IOFetchTest::prematureStopTest() header.
 TEST_F(IOFetchTest, UdpPrematureStop) {
-    expected_ = IOFetch::STOPPED;
-
-    // Stop before it is started
-    udp_fetch_.stop();
-    service_.get_io_service().post(udp_fetch_);
-
-    service_.run();
-    EXPECT_TRUE(run_);
+    prematureStopTest(IOFetch::UDP, udp_fetch_);
 }
 
-// Test that it will timeout when no answer arrives.
+// UDP premature stop test - see IOFetchTest::timeoutTest() header.
 TEST_F(IOFetchTest, UdpTimeout) {
-    expected_ = IOFetch::TIME_OUT;
-
-    service_.get_io_service().post(udp_fetch_);
-    service_.run();
-    EXPECT_TRUE(run_);
+    timeoutTest(IOFetch::UDP, udp_fetch_);
 }
 
-// Test that it will succeed when we fake an answer and stores the same data we
-// send.  This is done through a real socket on the loopback address.
-TEST_F(IOFetchTest, UdpReceive) {
+// UDP SendReceive test.  Set up a UDP server then ports a UDP fetch object.
+// This will send question_ to the server and receive the answer back from it.
+TEST_F(IOFetchTest, UdpSendReceive) {
+    protocol_ = IOFetch::UDP;
     expected_ = IOFetch::SUCCESS;
 
+    // Set up the server.
     udp::socket socket(service_.get_io_service(), udp::v4());
     socket.set_option(socket_base::reuse_address(true));
     socket.bind(udp::endpoint(TEST_HOST, TEST_PORT));
+    return_data_ = "Message returned to the client";
 
     udp::endpoint remote;
-    socket.async_receive_from(asio::buffer(server_buff_),
+    socket.async_receive_from(asio::buffer(receive_buffer_, sizeof(receive_buffer_)),
         remote,
-        boost::bind(&IOFetchTest::respond, this, &remote, &socket, _1, _2));
+        boost::bind(&IOFetchTest::udpReceiveHandler, this, &remote, &socket,
+                    _1, _2));
     service_.get_io_service().post(udp_fetch_);
+    if (debug_) {
+        cout << "udpSendReceive: async_receive_from posted, waiting for callback" <<
+                endl;
+    }
     service_.run();
 
     socket.close();
 
-    EXPECT_TRUE(run_);
-    ASSERT_EQ(sizeof TEST_DATA, buff_->getLength());
-    EXPECT_EQ(0, memcmp(TEST_DATA, buff_->getData(), sizeof TEST_DATA));
+    EXPECT_TRUE(run_);;
+}
+
+// Do the same tests for TCP transport
+
+TEST_F(IOFetchTest, TcpStop) {
+    stopTest(IOFetch::TCP, tcp_fetch_);
+}
+
+TEST_F(IOFetchTest, TcpPrematureStop) {
+    prematureStopTest(IOFetch::TCP, tcp_fetch_);
+}
+
+TEST_F(IOFetchTest, TcpTimeout) {
+    timeoutTest(IOFetch::TCP, tcp_fetch_);
+}
+
+// Test with values at or near 0, then at or near the chunk size (16 and 32
+// bytes, the sizes of the first two packets) then up to 65535.  These are done
+// in separate tests because in practice a new IOFetch is created for each
+// query/response exchange and we don't want to confuse matters in the test
+// by running the test with an IOFetch that has already done one exchange.
+
+TEST_F(IOFetchTest, TcpSendReceive0) {
+    tcpSendReturnTest(test_data_.substr(0, 0));
+}
+
+TEST_F(IOFetchTest, TcpSendReceive1) {
+    tcpSendReturnTest(test_data_.substr(0, 1));
+}
+
+TEST_F(IOFetchTest, TcpSendReceive15) {
+    tcpSendReturnTest(test_data_.substr(0, 15));
+}
+
+TEST_F(IOFetchTest, TcpSendReceive16) {
+    tcpSendReturnTest(test_data_.substr(0, 16));
+}
+
+TEST_F(IOFetchTest, TcpSendReceive17) {
+    tcpSendReturnTest(test_data_.substr(0, 17));
+}
+
+TEST_F(IOFetchTest, TcpSendReceive31) {
+    tcpSendReturnTest(test_data_.substr(0, 31));
+}
+
+TEST_F(IOFetchTest, TcpSendReceive32) {
+    tcpSendReturnTest(test_data_.substr(0, 32));
+}
+
+TEST_F(IOFetchTest, TcpSendReceive33) {
+    tcpSendReturnTest(test_data_.substr(0, 33));
+}
+
+TEST_F(IOFetchTest, TcpSendReceive4096) {
+    tcpSendReturnTest(test_data_.substr(0, 4096));
+}
+
+TEST_F(IOFetchTest, TcpSendReceive8192) {
+    tcpSendReturnTest(test_data_.substr(0, 8192));
+}
+
+TEST_F(IOFetchTest, TcpSendReceive16384) {
+    tcpSendReturnTest(test_data_.substr(0, 16384));
+}
+
+TEST_F(IOFetchTest, TcpSendReceive32768) {
+    tcpSendReturnTest(test_data_.substr(0, 32768));
+}
+
+TEST_F(IOFetchTest, TcpSendReceive65535) {
+    tcpSendReturnTest(test_data_.substr(0, 65535));
 }
 
 } // namespace asiolink

src/lib/asiolink/tests/run_unittests.cc

@@ -14,13 +14,15 @@
 
 #include <gtest/gtest.h>
 
+#include <log/root_logger_name.h>
 #include <dns/tests/unittest_util.h>
 
 int
 main(int argc, char* argv[])
 {
-    ::testing::InitGoogleTest(&argc, argv);
-    isc::UnitTestUtil::addDataPath(TEST_DATA_DIR);
+    ::testing::InitGoogleTest(&argc, argv);         // Initialize Google test
+    isc::log::setRootLoggerName("unittest");        // Set a root logger name
+    isc::UnitTestUtil::addDataPath(TEST_DATA_DIR);  // Add location of test data
 
     return (RUN_ALL_TESTS());
 }

src/lib/asiolink/tests/tcp_endpoint_unittest.cc

@@ -0,0 +1,55 @@
+// Copyright (C) 2011  Internet Systems Consortium, Inc. ("ISC")
+//
+// Permission to use, copy, modify, and/or distribute this software for any
+// purpose with or without fee is hereby granted, provided that the above
+// copyright notice and this permission notice appear in all copies.
+//
+// THE SOFTWARE IS PROVIDED "AS IS" AND ISC DISCLAIMS ALL WARRANTIES WITH
+// REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
+// AND FITNESS.  IN NO EVENT SHALL ISC BE LIABLE FOR ANY SPECIAL, DIRECT,
+// INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
+// LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
+// OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
+// PERFORMANCE OF THIS SOFTWARE.
+
+#include <config.h>
+
+#include <string>
+
+#include <gtest/gtest.h>
+
+#include <asio.hpp>
+#include <asiolink/io_address.h>
+#include <asiolink/tcp_endpoint.h>
+
+using namespace asiolink;
+using namespace std;
+
+// This test checks that the endpoint can manage its own internal
+// asio::ip::tcp::endpoint object.
+
+TEST(TCPEndpointTest, v4Address) {
+    const string test_address("192.0.2.1");
+    const unsigned short test_port = 5301;
+
+    IOAddress address(test_address);
+    TCPEndpoint endpoint(address, test_port);
+
+    EXPECT_TRUE(address == endpoint.getAddress());
+    EXPECT_EQ(test_port, endpoint.getPort());
+    EXPECT_EQ(IPPROTO_TCP, endpoint.getProtocol());
+    EXPECT_EQ(AF_INET, endpoint.getFamily());
+}
+
+TEST(TCPEndpointTest, v6Address) {
+    const string test_address("2001:db8::1235");
+    const unsigned short test_port = 5302;
+
+    IOAddress address(test_address);
+    TCPEndpoint endpoint(address, test_port);
+
+    EXPECT_TRUE(address == endpoint.getAddress());
+    EXPECT_EQ(test_port, endpoint.getPort());
+    EXPECT_EQ(IPPROTO_TCP, endpoint.getProtocol());
+    EXPECT_EQ(AF_INET6, endpoint.getFamily());
+}

src/lib/asiolink/tests/tcp_socket_unittest.cc

@@ -0,0 +1,515 @@
+// Copyright (C) 2011  Internet Systems Consortium, Inc. ("ISC")
+//
+// Permission to use, copy, modify, and/or distribute this software for any
+// purpose with or without fee is hereby granted, provided that the above
+// copyright notice and this permission notice appear in all copies.
+//
+// THE SOFTWARE IS PROVIDED "AS IS" AND ISC DISCLAIMS ALL WARRANTIES WITH
+// REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
+// AND FITNESS.  IN NO EVENT SHALL ISC BE LIABLE FOR ANY SPECIAL, DIRECT,
+// INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
+// LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
+// OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
+// PERFORMANCE OF THIS SOFTWARE.
+
+/// \brief Test of TCPSocket
+///
+/// Tests the fuctionality of a TCPSocket by working through an open-send-
+/// receive-close sequence and checking that the asynchronous notifications
+/// work.
+
+#include <string>
+
+#include <arpa/inet.h>
+#include <netinet/in.h>
+#include <sys/types.h>
+#include <sys/socket.h>
+
+#include <algorithm>
+#include <cstdlib>
+#include <cstddef>
+#include <vector>
+
+#include <gtest/gtest.h>
+
+#include <boost/bind.hpp>
+#include <boost/shared_ptr.hpp>
+
+#include <dns/buffer.h>
+
+#include <asio.hpp>
+
+#include <asiolink/asiolink_utilities.h>
+#include <asiolink/io_service.h>
+#include <asiolink/tcp_endpoint.h>
+#include <asiolink/tcp_socket.h>
+
+using namespace asio;
+using namespace asio::ip;
+using namespace asiolink;
+using namespace isc::dns;
+using namespace std;
+
+namespace {
+
+const char SERVER_ADDRESS[] = "127.0.0.1";
+const unsigned short SERVER_PORT = 5303;
+
+// TODO: Shouldn't we send something that is real message?
+const char OUTBOUND_DATA[] = "Data sent from client to server";
+const char INBOUND_DATA[] = "Returned data from server to client";
+}
+
+/// An instance of this object is passed to the asynchronous I/O functions
+/// and the operator() method is called when when an asynchronous I/O completes.
+/// The arguments to the completion callback are stored for later retrieval.
+class TCPCallback {
+public:
+    /// \brief Operations the server is doing
+    enum Operation {
+        ACCEPT = 0,     ///< accept() was issued
+        OPEN = 1,       /// Client connected to server
+        READ = 2,       ///< Asynchronous read completed
+        WRITE = 3,      ///< Asynchronous write completed
+        NONE = 4        ///< "Not set" state
+    };
+
+    /// \brief Minimim size of buffers
+    enum {
+        MIN_SIZE = (64 * 1024 + 2)          ///< 64kB + two bytes for a count
+    };
+
+    struct PrivateData {
+        PrivateData() :
+            error_code_(), length_(0), cumulative_(0), expected_(0), offset_(0),
+            name_(""), queued_(NONE), called_(NONE)
+        {}
+
+        asio::error_code    error_code_;    ///< Completion error code
+        size_t              length_;        ///< Bytes transferred in this I/O
+        size_t              cumulative_;    ///< Cumulative bytes transferred
+        size_t              expected_;      ///< Expected amount of data
+        size_t              offset_;        ///< Where to put data in buffer
+        std::string         name_;          ///< Which of the objects this is
+        Operation           queued_;        ///< Queued operation
+        Operation           called_;        ///< Which callback called
+        uint8_t             data_[MIN_SIZE];  ///< Receive buffer
+
+    };
+
+    /// \brief Constructor
+    ///
+    /// Constructs the object.  It also creates the data member pointed to by
+    /// a shared pointer.  When used as a callback object, this is copied as it
+    /// is passed into the asynchronous function.  This means that there are two
+    /// objects and inspecting the one we passed in does not tell us anything.
+    ///
+    /// Therefore we use a boost::shared_ptr.  When the object is copied, the
+    /// shared pointer is copied, which leaves both objects pointing to the same
+    /// data.
+    ///
+    /// \param which Which of the two callback objects this is
+    TCPCallback(std::string which) : ptr_(new PrivateData())
+    {
+        ptr_->name_ = which;
+    }
+
+    /// \brief Destructor
+    ///
+    /// No code needed, destroying the shared pointer destroys the private data.
+    virtual ~TCPCallback()
+    {}
+
+    /// \brief Client Callback Function
+    ///
+    /// Called when an asynchronous operation is completed by the client, this
+    /// stores the origin of the operation in the client_called_ data member.
+    ///
+    /// \param ec I/O completion error code passed to callback function.
+    /// \param length Number of bytes transferred
+    void operator()(asio::error_code ec = asio::error_code(),
+                            size_t length = 0)
+    {
+        setCode(ec.value());
+        ptr_->called_ = ptr_->queued_;
+        ptr_->length_ = length;
+    }
+
+    /// \brief Get I/O completion error code
+    int getCode() {
+        return (ptr_->error_code_.value());
+    }
+
+    /// \brief Set I/O completion code
+    ///
+    /// \param code New value of completion code
+    void setCode(int code) {
+        ptr_->error_code_ = asio::error_code(code, asio::error_code().category());
+    }
+
+    /// \brief Get number of bytes transferred in I/O
+    size_t& length() {
+        return (ptr_->length_);
+    }
+
+    /// \brief Get cumulative number of bytes transferred in I/O
+    size_t& cumulative() {
+        return (ptr_->cumulative_);
+    }
+
+    /// \brief Get expected amount of data
+    size_t& expected() {
+        return (ptr_->expected_);
+    }
+
+    /// \brief Get offset intodData
+    size_t& offset() {
+        return (ptr_->offset_);
+    }
+
+    /// \brief Get data member
+    uint8_t* data() {
+        return (ptr_->data_);
+    }
+
+    /// \brief Get flag to say what was queued
+    Operation& queued() {
+        return (ptr_->queued_);
+    }
+
+    /// \brief Get flag to say when callback was called
+    Operation& called() {
+        return (ptr_->called_);
+    }
+
+    /// \brief Return instance of callback name
+    std::string& name() {
+        return (ptr_->name_);
+    }
+
+private:
+    boost::shared_ptr<PrivateData>  ptr_;   ///< Pointer to private data
+};
+
+
+// Read Server Data
+//
+// Called in the part of the test that has the client send a message to the
+// server, this loops until all the data has been read (synchronously) by the
+// server.
+//
+// "All the data read" means that the server has received a message that is
+// preceded by a two-byte count field and that the total amount of data received
+// from the remote end is equal to the value in the count field plus two bytes
+// for the count field itself.
+//
+// \param socket Socket on which the server is reading data
+// \param server_cb Structure in which server data is held.
+void
+serverRead(tcp::socket& socket, TCPCallback& server_cb) {
+
+    // As we may need to read multiple times, keep a count of the cumulative
+    // amount of data read and do successive reads into the appropriate part
+    // of the buffer.
+    //
+    // Note that there are no checks for buffer overflow - this is a test
+    // program and we have sized the buffer to be large enough for the test.
+    server_cb.cumulative() = 0;
+
+    bool complete = false;
+    while (!complete) {
+
+        // Read block of data and update cumulative amount of data received.
+        server_cb.length() = socket.receive(
+            asio::buffer(server_cb.data() + server_cb.cumulative(),
+                TCPCallback::MIN_SIZE - server_cb.cumulative()));
+        server_cb.cumulative() += server_cb.length();
+
+        // If we have read at least two bytes, we can work out how much we
+        // should be reading.
+        if (server_cb.cumulative() >= 2) {
+           server_cb.expected() = readUint16(server_cb.data());
+            if ((server_cb.expected() + 2) == server_cb.cumulative()) {
+
+                // Amount of data read from socket equals the size of the
+                // message (as indicated in the first two bytes of the message)
+                // plus the size of the count field.  Therefore we have received
+                // all the data.
+                complete = true;
+            }
+        }
+    }
+}
+
+// Receive complete method should return true only if the count in the first
+// two bytes is equal to the size of the rest if the buffer.
+
+TEST(TCPSocket, processReceivedData) {
+    const uint16_t PACKET_SIZE = 16382;     // Amount of "real" data in the buffer
+
+    IOService               service;        // Used to instantiate socket
+    TCPSocket<TCPCallback>  test(service);  // Socket under test
+    uint8_t                 inbuff[PACKET_SIZE + 2];   // Buffer to check
+    OutputBufferPtr         outbuff(new OutputBuffer(16));
+                                            // Where data is put
+    size_t                  expected;       // Expected amount of data
+    size_t                  offset;         // Where to put next data
+    size_t                  cumulative;     // Cumulative data received
+
+    // Set some dummy values in the buffer to check
+    for (size_t i = 0; i < sizeof(inbuff); ++i) {
+        inbuff[i] = i % 256;
+    }
+
+    // Check that the method will handle various receive sizes.
+    writeUint16(PACKET_SIZE, inbuff);
+
+    cumulative = 0;
+    offset = 0;
+    expected = 0;
+    outbuff->clear();
+    bool complete = test.processReceivedData(inbuff, 1, cumulative, offset,
+                                             expected, outbuff);
+    EXPECT_FALSE(complete);
+    EXPECT_EQ(1, cumulative);
+    EXPECT_EQ(1, offset);
+    EXPECT_EQ(0, expected);
+    EXPECT_EQ(0, outbuff->getLength());
+
+    // Now pretend that we've received one more byte.
+    complete = test.processReceivedData(inbuff, 1, cumulative, offset, expected,
+                                        outbuff);
+    EXPECT_FALSE(complete);
+    EXPECT_EQ(2, cumulative);
+    EXPECT_EQ(0, offset);
+    EXPECT_EQ(PACKET_SIZE, expected);
+    EXPECT_EQ(0, outbuff->getLength());
+
+    // Add another two bytes.  However, this time note that we have to offset
+    // in the input buffer because it is expected that the next chunk of data
+    // from the connection will be read into the start of the buffer.
+    complete = test.processReceivedData(inbuff + cumulative, 2, cumulative,
+                                        offset, expected, outbuff);
+    EXPECT_FALSE(complete);
+    EXPECT_EQ(4, cumulative);
+    EXPECT_EQ(0, offset);
+    EXPECT_EQ(PACKET_SIZE, expected);
+    EXPECT_EQ(2, outbuff->getLength());
+
+    const uint8_t* dataptr = static_cast<const uint8_t*>(outbuff->getData());
+    EXPECT_TRUE(equal(inbuff + 2, inbuff + cumulative, dataptr));
+
+    // And add the remaining data.  Remember that "inbuff" is "PACKET_SIZE + 2"
+    // long.
+    complete = test.processReceivedData(inbuff + cumulative,
+                                        PACKET_SIZE + 2 - cumulative,
+                                        cumulative, offset, expected, outbuff);
+    EXPECT_TRUE(complete);
+    EXPECT_EQ(PACKET_SIZE + 2, cumulative);
+    EXPECT_EQ(0, offset);
+    EXPECT_EQ(PACKET_SIZE, expected);
+    EXPECT_EQ(PACKET_SIZE, outbuff->getLength());
+    dataptr = static_cast<const uint8_t*>(outbuff->getData());
+    EXPECT_TRUE(equal(inbuff + 2, inbuff + cumulative, dataptr));
+}
+
+// TODO: Need to add a test to check the cancel() method
+
+// Tests the operation of a TCPSocket by opening it, sending an asynchronous
+// message to a server, receiving an asynchronous message from the server and
+// closing.
+TEST(TCPSocket, SequenceTest) {
+
+    // Common objects.
+    IOService   service;                    // Service object for async control
+
+    // The client - the TCPSocket being tested
+    TCPSocket<TCPCallback>  client(service);// Socket under test
+    TCPCallback client_cb("Client");        // Async I/O callback function
+    TCPEndpoint client_remote_endpoint;     // Where client receives message from
+    OutputBufferPtr client_buffer(new OutputBuffer(128));
+                                            // Received data is put here
+
+    // The server - with which the client communicates.
+    IOAddress   server_address(SERVER_ADDRESS);
+                                            // Address of target server
+    TCPCallback server_cb("Server");        // Server callback
+    TCPEndpoint server_endpoint(server_address, SERVER_PORT);
+                                            // Endpoint describing server
+    TCPEndpoint server_remote_endpoint;     // Address where server received message from
+    tcp::socket server_socket(service.get_io_service());
+                                            // Socket used for server
+
+    // Step 1.  Create the connection between the client and the server.  Set
+    // up the server to accept incoming connections and have the client open
+    // a channel to it.
+
+    // Set up server - open socket and queue an accept.
+    server_cb.queued() = TCPCallback::ACCEPT;
+    server_cb.called() = TCPCallback::NONE;
+    server_cb.setCode(42);  // Some error
+    tcp::acceptor acceptor(service.get_io_service(),
+                            tcp::endpoint(tcp::v4(), SERVER_PORT));
+    acceptor.set_option(tcp::acceptor::reuse_address(true));
+    acceptor.async_accept(server_socket, server_cb);
+
+    // Set up client - connect to the server.
+    client_cb.queued() = TCPCallback::OPEN;
+    client_cb.called() = TCPCallback::NONE;
+    client_cb.setCode(43);  // Some error
+    EXPECT_FALSE(client.isOpenSynchronous());
+    client.open(&server_endpoint, client_cb);
+
+    // Run the open and the accept callback and check that they ran.
+    service.run_one();
+    service.run_one();
+
+    EXPECT_EQ(TCPCallback::ACCEPT, server_cb.called());
+    EXPECT_EQ(0, server_cb.getCode());
+
+    EXPECT_EQ(TCPCallback::OPEN, client_cb.called());
+    EXPECT_EQ(0, client_cb.getCode());
+
+    // Step 2.  Get the client to write to the server asynchronously.  The
+    // server will loop reading the data synchronously.
+
+    // Write asynchronously to the server.
+    client_cb.called() = TCPCallback::NONE;
+    client_cb.queued() = TCPCallback::WRITE;
+    client_cb.setCode(143);  // Arbitrary number
+    client_cb.length() = 0;
+    client.asyncSend(OUTBOUND_DATA, sizeof(OUTBOUND_DATA), &server_endpoint, client_cb);
+
+    // Wait for the client callback to complete. (Must do this first on
+    // Solaris: if we do the synchronous read first, the test hangs.)
+    service.run_one();
+
+    // Synchronously read the data from the server.;
+    serverRead(server_socket, server_cb);
+
+    // Check the client state
+    EXPECT_EQ(TCPCallback::WRITE, client_cb.called());
+    EXPECT_EQ(0, client_cb.getCode());
+    EXPECT_EQ(sizeof(OUTBOUND_DATA) + 2, client_cb.length());
+
+    // ... and check what the server received.
+    EXPECT_EQ(sizeof(OUTBOUND_DATA) + 2, server_cb.cumulative());
+    EXPECT_TRUE(equal(OUTBOUND_DATA,
+                (OUTBOUND_DATA + (sizeof(OUTBOUND_DATA) - 1)),
+                (server_cb.data() + 2)));
+
+    // Step 3.  Get the server to write all the data asynchronously and have the
+    // client loop (asynchronously) reading the data.  Note that we copy the
+    // data into the server's internal buffer in order to precede it with a two-
+    // byte count field.
+
+    // Have the server write asynchronously to the client.
+    server_cb.called() = TCPCallback::NONE;
+    server_cb.queued() = TCPCallback::WRITE;
+    server_cb.length() = 0;
+    server_cb.cumulative() = 0;
+
+    writeUint16(sizeof(INBOUND_DATA), server_cb.data());
+    copy(INBOUND_DATA, (INBOUND_DATA + sizeof(INBOUND_DATA) - 1),
+        (server_cb.data() + 2));
+    server_socket.async_send(asio::buffer(server_cb.data(),
+                                          (sizeof(INBOUND_DATA) + 2)),
+                             server_cb);
+
+    // Have the client read asynchronously.
+    client_cb.called() = TCPCallback::NONE;
+    client_cb.queued() = TCPCallback::READ;
+    client_cb.length() = 0;
+    client_cb.cumulative() = 0;
+    client_cb.expected() = 0;
+    client_cb.offset() = 0;
+
+    client.asyncReceive(client_cb.data(), TCPCallback::MIN_SIZE,
+                        client_cb.offset(), &client_remote_endpoint,
+                        client_cb);
+
+    // Run the callbacks. Several options are possible depending on how ASIO
+    // is implemented and whether the message gets fragmented:
+    //
+    // 1) The send handler may complete immediately, regardess of whether the
+    // data has been read by the client.  (This is the most likely.)
+    // 2) The send handler may only run after all the data has been read by
+    // the client. (This could happen if the client's TCP buffers were too
+    // small so the data was not transferred to the "remote" system until the
+    // remote buffer has been emptied one or more times.)
+    // 3) The client handler may be run a number of times to handle the message
+    // fragments and the server handler may run between calls of the client
+    // handler.
+    //
+    // So loop, running one handler at a time until we are certain that all the
+    // handlers have run.
+
+    bool server_complete = false;
+    bool client_complete = false;
+    while (!server_complete || !client_complete) {
+        service.run_one();
+
+        // Has the server run?
+        if (!server_complete) {
+            if (server_cb.called() == server_cb.queued()) {
+
+                // Yes.  Check that the send completed successfully and that
+                // all the data that was expected to have been sent was in fact
+                // sent.
+                EXPECT_EQ(0, server_cb.getCode());
+                EXPECT_EQ((sizeof(INBOUND_DATA) + 2), server_cb.length());
+                server_complete = true;
+                continue;
+            }
+        }
+
+        if (!client_complete) {
+
+            // Client callback must have run.  Check that it ran OK.
+            EXPECT_EQ(TCPCallback::READ, client_cb.called());
+            EXPECT_EQ(0, client_cb.getCode());
+
+            // Check if we need to queue another read, copying the data into
+            // the output buffer as we do so.
+            client_complete = client.processReceivedData(client_cb.data(),
+                                                         client_cb.length(),
+                                                         client_cb.cumulative(),
+                                                         client_cb.offset(),
+                                                         client_cb.expected(),
+                                                         client_buffer);
+
+            // If the data is not complete, queue another read.
+            if (! client_complete) {
+                client_cb.called() = TCPCallback::NONE;
+                client_cb.queued() = TCPCallback::READ;
+                client_cb.length() = 0;
+                client.asyncReceive(client_cb.data(), TCPCallback::MIN_SIZE ,
+                                    client_cb.offset(), &client_remote_endpoint,
+                                    client_cb);
+            }
+        }
+    }
+
+    // Both the send and the receive have completed.  Check that the received
+    // is what was sent.
+
+    // Check the client state
+    EXPECT_EQ(TCPCallback::READ, client_cb.called());
+    EXPECT_EQ(0, client_cb.getCode());
+    EXPECT_EQ(sizeof(INBOUND_DATA) + 2, client_cb.cumulative());
+    EXPECT_EQ(sizeof(INBOUND_DATA), client_buffer->getLength());
+
+    // ... and check what the server sent.
+    EXPECT_EQ(TCPCallback::WRITE, server_cb.called());
+    EXPECT_EQ(0, server_cb.getCode());
+    EXPECT_EQ(sizeof(INBOUND_DATA) + 2, server_cb.length());
+
+    // ... and that what was sent is what was received.
+    const uint8_t* received = static_cast<const uint8_t*>(client_buffer->getData());
+    EXPECT_TRUE(equal(INBOUND_DATA, (INBOUND_DATA + (sizeof(INBOUND_DATA) - 1)),
+                      received));
+
+    // Close client and server.
+    EXPECT_NO_THROW(client.close());
+    EXPECT_NO_THROW(server_socket.close());
+}

src/lib/asiolink/tests/udp_socket_unittest.cc

@@ -12,21 +12,6 @@
 // OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
 // PERFORMANCE OF THIS SOFTWARE.
 
-// Copyright (C) 2011  Internet Systems Consortium, Inc. ("ISC")
-//
-// Permission to use, copy, modify, and/or distribute this software for any
-// purpose with or without fee is hereby granted, provided that the above
-// copyright notice and this permission notice appear in all copies.
-//
-// THE SOFTWARE IS PROVIDED "AS IS" AND ISC DISCLAIMS ALL WARRANTIES WITH
-// REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
-// AND FITNESS.  IN NO EVENT SHALL ISC BE LIABLE FOR ANY SPECIAL, DIRECT,
-// INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
-// LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
-// OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
-// PERFORMANCE OF THIS SOFTWARE.
-
-
 /// \brief Test of UDPSocket
 ///
 /// Tests the fuctionality of a UDPSocket by working through an open-send-
@@ -50,14 +35,18 @@
 #include <boost/bind.hpp>
 #include <boost/shared_ptr.hpp>
 
+#include <dns/buffer.h>
+
 #include <asio.hpp>
 
+#include <asiolink/asiolink_utilities.h>
 #include <asiolink/io_service.h>
 #include <asiolink/udp_endpoint.h>
 #include <asiolink/udp_socket.h>
 
 using namespace asio;
 using namespace asiolink;
+using namespace isc::dns;
 using namespace std;
 
 namespace {
@@ -177,6 +166,49 @@ private:
     boost::shared_ptr<PrivateData>  ptr_;   ///< Pointer to private data
 };
 
+// Receive complete method should return true regardless of what is in the first
+// two bytes of a buffer.
+
+TEST(UDPSocket, processReceivedData) {
+    IOService               service;        // Used to instantiate socket
+    UDPSocket<UDPCallback>  test(service);  // Socket under test
+    uint8_t                 inbuff[32];     // Buffer to check
+    OutputBufferPtr         outbuff(new OutputBuffer(16));
+                                            // Where data is put
+    size_t                  expected;       // Expected amount of data
+    size_t                  offset;         // Where to put next data
+    size_t                  cumulative;     // Cumulative data received
+
+    // Set some dummy values in the buffer to check
+    for (uint8_t i = 0; i < sizeof(inbuff); ++i) {
+        inbuff[i] = i;
+    }
+
+    // Expect that the value is true whatever number is written in the first
+    // two bytes of the buffer.
+    uint16_t count = 0;
+    for (uint32_t i = 0; i < (2 << 16); ++i, ++count) {
+        writeUint16(count, inbuff);
+
+        // Set some random values
+        cumulative = 5;
+        offset = 10;
+        expected = 15;
+        outbuff->clear();
+
+        bool completed = test.processReceivedData(inbuff, sizeof(inbuff),
+                                                  cumulative, offset, expected,
+                                                  outbuff);
+        EXPECT_TRUE(completed);
+        EXPECT_EQ(sizeof(inbuff), cumulative);
+        EXPECT_EQ(0, offset);
+        EXPECT_EQ(sizeof(inbuff), expected);
+
+        const uint8_t* dataptr = static_cast<const uint8_t*>(outbuff->getData());
+        EXPECT_TRUE(equal(inbuff, inbuff + sizeof(inbuff) - 1, dataptr));
+    }
+}
+
 // TODO: Need to add a test to check the cancel() method
 
 // Tests the operation of a UDPSocket by opening it, sending an asynchronous
@@ -199,6 +231,10 @@ TEST(UDPSocket, SequenceTest) {
     UDPCallback client_cb("Client");        // Async I/O callback function
     UDPEndpoint client_remote_endpoint;     // Where client receives message from
     size_t      client_cumulative = 0;      // Cumulative data received
+    size_t      client_offset = 0;          // Offset into buffer where data is put
+    size_t      client_expected = 0;        // Expected amount of data
+    OutputBufferPtr client_buffer(new OutputBuffer(16));
+                                            // Where data is put
 
     // The server - with which the client communicates.  For convenience, we
     // use the same io_service, and use the endpoint object created for
@@ -208,11 +244,12 @@ TEST(UDPSocket, SequenceTest) {
     server.set_option(socket_base::reuse_address(true));
 
     // Assertion to ensure that the server buffer is large enough
-    char data[UDPSocket<UDPCallback>::MAX_SIZE];
+    char data[UDPSocket<UDPCallback>::MIN_SIZE];
     ASSERT_GT(sizeof(data), sizeof(OUTBOUND_DATA));
 
     // Open the client socket - the operation should be synchronous
-    EXPECT_FALSE(client.open(&server_endpoint, client_cb));
+    EXPECT_TRUE(client.isOpenSynchronous());
+    client.open(&server_endpoint, client_cb);
 
     // Issue read on the server.  Completion callback should not have run.
     server_cb.setCalled(false);
@@ -257,7 +294,7 @@ TEST(UDPSocket, SequenceTest) {
     server.async_send_to(buffer(INBOUND_DATA, sizeof(INBOUND_DATA)),
         server_remote_endpoint.getASIOEndpoint(), server_cb);
 
-    // Expect the two callbacks to run
+    // Expect two callbacks to run.
     service.run_one();
     service.run_one();
 
@@ -276,10 +313,19 @@ TEST(UDPSocket, SequenceTest) {
     EXPECT_TRUE(server_address == client_remote_endpoint.getAddress());
     EXPECT_EQ(SERVER_PORT, client_remote_endpoint.getPort());
 
-    // Finally, check that the receive received a complete buffer's worth of data.
-    EXPECT_TRUE(client.receiveComplete(&data[0], client_cb.getLength(),
-        client_cumulative));
+    // Check that the receive received a complete buffer's worth of data.
+    EXPECT_TRUE(client.processReceivedData(&data[0], client_cb.getLength(),
+                                           client_cumulative, client_offset,
+                                           client_expected, client_buffer));
+
     EXPECT_EQ(client_cb.getLength(), client_cumulative);
+    EXPECT_EQ(0, client_offset);
+    EXPECT_EQ(client_cb.getLength(), client_expected);
+    EXPECT_EQ(client_cb.getLength(), client_buffer->getLength());
+
+    // ...and check that the data was copied to the output client buffer.
+    const char* client_char_data = static_cast<const char*>(client_buffer->getData());
+    EXPECT_TRUE(equal(&data[0], &data[client_cb.getLength() - 1], client_char_data));
 
     // Close client and server.
     EXPECT_NO_THROW(client.close());

src/lib/asiolink/udp_endpoint.h

@@ -64,6 +64,17 @@ public:
         asio_endpoint_placeholder_(NULL), asio_endpoint_(asio_endpoint)
     {}
 
+    /// \brief Constructor from an ASIO UDP endpoint.
+    ///
+    /// This constructor is designed to be an efficient wrapper for the
+    /// corresponding ASIO class, \c udp::endpoint.
+    ///
+    /// \param asio_endpoint The ASIO representation of the TCP endpoint.
+    UDPEndpoint(const asio::ip::udp::endpoint& asio_endpoint) :
+        asio_endpoint_placeholder_(new asio::ip::udp::endpoint(asio_endpoint)),
+        asio_endpoint_(*asio_endpoint_placeholder_)
+    {}
+
     /// \brief The destructor.
     virtual ~UDPEndpoint() { delete asio_endpoint_placeholder_; }
     //@}

src/lib/asiolink/udp_server.cc

@@ -15,6 +15,7 @@
 #include <netinet/in.h>
 #include <sys/socket.h>
 #include <unistd.h>             // for some IPC/network system calls
+#include <errno.h>
 
 #include <boost/shared_array.hpp>
 
@@ -188,12 +189,12 @@ UDPServer::operator()(error_code ec, size_t length) {
                     buffer(data_->data_.get(), MAX_LENGTH), *data_->sender_,
                     *this);
 
-                //return if we met fatal error
-                //Todo add log
+                // Abort on fatal errors
+                // TODO: add log
                 if (ec) {
                     using namespace asio::error;
                     if (ec.value() != would_block && ec.value() != try_again &&
-                            ec.value() != interrupted) {
+                        ec.value() != interrupted) {
                         return;
                     }
                 }
@@ -260,8 +261,6 @@ UDPServer::operator()(error_code ec, size_t length) {
         // this point.
         CORO_YIELD data_->io_.post(AsyncLookup<UDPServer>(*this));
 
-        dlog("[XX] got an answer");
-
         // The 'done_' flag indicates whether we have an answer
         // to send back.  If not, exit the coroutine permanently.
         if (!data_->done_) {

src/lib/asiolink/udp_socket.h

@@ -28,7 +28,6 @@
 
 #include <config.h>
 
-
 #include <asiolink/io_asio_socket.h>
 #include <asiolink/io_endpoint.h>
 #include <asiolink/io_service.h>
@@ -49,20 +48,20 @@ private:
 
 public:
     enum {
-        MAX_SIZE = 4096         // Send and receive size
+        MIN_SIZE = 4096         // Minimum send and receive size
     };
-    
+
     /// \brief Constructor from an ASIO UDP socket.
     ///
-    /// \param socket The ASIO representation of the UDP socket.  It
-    /// is assumed that the caller will open and close the socket, so
-    /// these operations are a no-op for that socket.
+    /// \param socket The ASIO representation of the UDP socket.  It is assumed
+    ///        that the caller will open and close the socket, so these
+    ///        operations are a no-op for that socket.
     UDPSocket(asio::ip::udp::socket& socket);
 
     /// \brief Constructor
     ///
     /// Used when the UDPSocket is being asked to manage its own internal
-    /// socket.  It is assumed that open() and close() will not be used.
+    /// socket.  In this case, the open() and close() methods are used.
     ///
     /// \param service I/O Service object used to manage the socket.
     UDPSocket(IOService& service);
@@ -70,68 +69,79 @@ public:
     /// \brief Destructor
     virtual ~UDPSocket();
 
-    virtual int getNative() const { return (socket_.native()); }
-    virtual int getProtocol() const { return (IPPROTO_UDP); }
+    /// \brief Return file descriptor of underlying socket
+    virtual int getNative() const {
+        return (socket_.native());
+    }
 
-    /// \brief Open Socket
+    /// \brief Return protocol of socket
+    virtual int getProtocol() const {
+        return (IPPROTO_UDP);
+    }
+
+    /// \brief Is "open()" synchronous?
     ///
-    /// Opens the UDP socket.  In the model for transport-layer agnostic I/O,
-    /// an "open" operation includes a connection to the remote end (which
-    /// may take time).  This does not happen for UDP, so the method returns
-    /// "false" to indicate that the operation completed synchronously.
+    /// Indicates that the opening of a UDP socket is synchronous.
+    virtual bool isOpenSynchronous() const {
+        return true;
+    }
+
+    /// \brief Open Socket
     ///
-    /// \param endpoint Endpoint to which the socket will connect to.
-    /// \param callback Unused.
+    /// Opens the UDP socket.  This is a synchronous operation.
     ///
-    /// \return false to indicate that the "operation" completed synchronously.
-    virtual bool open(const IOEndpoint* endpoint, C&);
+    /// \param endpoint Endpoint to which the socket will send data.  This is
+    ///        used to determine the address family trhat should be used for the
+    ///        underlying socket.
+    /// \param callback Unused as the operation is synchronous.
+    virtual void open(const IOEndpoint* endpoint, C& callback);
 
     /// \brief Send Asynchronously
     ///
-    /// This corresponds to async_send_to() for UDP sockets and async_send()
-    /// for TCP.  In both cases an endpoint argument is supplied indicating the
-    /// target of the send - this is ignored for TCP.
+    /// Calls the underlying socket's async_send_to() method to send a packet of
+    /// data asynchronously to the remote endpoint.  The callback will be called
+    /// on completion.
     ///
     /// \param data Data to send
     /// \param length Length of data to send
     /// \param endpoint Target of the send
     /// \param callback Callback object.
     virtual void asyncSend(const void* data, size_t length,
-        const IOEndpoint* endpoint, C& callback);
+                           const IOEndpoint* endpoint, C& callback);
 
     /// \brief Receive Asynchronously
     ///
-    /// This correstponds to async_receive_from() for UDP sockets and
-    /// async_receive() for TCP.  In both cases, an endpoint argument is
-    /// supplied to receive the source of the communication.  For TCP it will
-    /// be filled in with details of the connection.
+    /// Calls the underlying socket's async_receive_from() method to read a
+    /// packet of data from a remote endpoint.  Arrival of the data is signalled
+    /// via a call to the callback function.
     ///
     /// \param data Buffer to receive incoming message
     /// \param length Length of the data buffer
-    /// \param cumulative Amount of data that should already be in the buffer.
-    /// (This is ignored - every UPD receive fills the buffer from the start.)
+    /// \param offset Offset into buffer where data is to be put
     /// \param endpoint Source of the communication
     /// \param callback Callback object
-    virtual void asyncReceive(void* data, size_t length, size_t cumulative,
-        IOEndpoint* endpoint, C& callback);
+    virtual void asyncReceive(void* data, size_t length, size_t offset,
+                              IOEndpoint* endpoint, C& callback);
 
-    /// \brief Checks if the data received is complete.
+    /// \brief Process received data
     ///
-    /// As all the data is received in one I/O, so this is, this is effectively
-    /// a no-op (although it does update the amount of data received).
+    /// See the description of IOAsioSocket::receiveComplete for a complete
+    /// description of this method.
     ///
-    /// \param data Data buffer containing data to date.  (This is ignored
-    /// for UDP receives.)
-    /// \param length Amount of data received in last asynchronous I/O
-    /// \param cumulative On input, amount of data received before the last
-    /// I/O.  On output, the total amount of data received to date.
+    /// \param staging Pointer to the start of the staging buffer.
+    /// \param length Amount of data in the staging buffer.
+    /// \param cumulative Amount of data received before the staging buffer is
+    ///        processed.
+    /// \param offset Unused.
+    /// \param expected unused.
+    /// \param outbuff Output buffer.  Data in the staging buffer is be copied
+    ///        to this output buffer in the call.
     ///
-    /// \return true if the receive is complete, false if another receive is
-    /// needed.
-    virtual bool receiveComplete(void*, size_t length, size_t& cumulative) {
-        cumulative = length;
-        return (true);
-    }
+    /// \return Always true
+    virtual bool processReceivedData(const void* staging, size_t length,
+                                     size_t& cumulative, size_t& offset,
+                                     size_t& expected,
+                                     isc::dns::OutputBufferPtr& outbuff);
 
     /// \brief Cancel I/O On Socket
     virtual void cancel();
@@ -174,16 +184,16 @@ UDPSocket<C>::~UDPSocket()
     delete socket_ptr_;
 }
 
-// Open the socket.  Throws an error on failure
-// TODO: Make the open more resilient
+// Open the socket.
 
-template <typename C> bool
+template <typename C> void
 UDPSocket<C>::open(const IOEndpoint* endpoint, C&) {
 
-    // Ignore opens on already-open socket.  Don't throw a failure because
-    // of uncertainties as to what precedes whan when using asynchronous I/O.
-    // At also allows us a treat a passed-in socket as a self-managed socket.
-
+    // Ignore opens on already-open socket.  (Don't throw a failure because
+    // of uncertainties as to what precedes whan when using asynchronous I/O.)
+    // It also allows us a treat a passed-in socket in exactly the same way as
+    // a self-managed socket (in that we can call the open() and close() methods
+    // of this class).
     if (!isopen_) {
         if (endpoint->getFamily() == AF_INET) {
             socket_.open(asio::ip::udp::v4());
@@ -193,14 +203,21 @@ UDPSocket<C>::open(const IOEndpoint* endpoint, C&) {
         }
         isopen_ = true;
 
-        // Ensure it can send and receive 4K buffers.
-        socket_.set_option(asio::socket_base::send_buffer_size(MAX_SIZE));
-        socket_.set_option(asio::socket_base::receive_buffer_size(MAX_SIZE));
-    ;
-        // Allow reuse of an existing port/address
-        socket_.set_option(asio::socket_base::reuse_address(true));
+        // Ensure it can send and receive at least 4K buffers.
+        asio::ip::udp::socket::send_buffer_size snd_size;
+        socket_.get_option(snd_size);
+        if (snd_size.value() < MIN_SIZE) {
+            snd_size = MIN_SIZE;
+            socket_.set_option(snd_size);
+        }
+
+        asio::ip::udp::socket::receive_buffer_size rcv_size;
+        socket_.get_option(rcv_size);
+        if (rcv_size.value() < MIN_SIZE) {
+            rcv_size = MIN_SIZE;
+            socket_.set_option(rcv_size);
+        }
     }
-    return (false);
 }
 
 // Send a message.  Should never do this if the socket is not open, so throw
@@ -208,19 +225,20 @@ UDPSocket<C>::open(const IOEndpoint* endpoint, C&) {
 
 template <typename C> void
 UDPSocket<C>::asyncSend(const void* data, size_t length,
-    const IOEndpoint* endpoint, C& callback)
+                        const IOEndpoint* endpoint, C& callback)
 {
     if (isopen_) {
 
         // Upconvert to a UDPEndpoint.  We need to do this because although
         // IOEndpoint is the base class of UDPEndpoint and TCPEndpoint, it
-        // doing cont contain a method for getting at the underlying endpoint
-        // type - those are in the derived class and the two classes differ on
+        // does not contain a method for getting at the underlying endpoint
+        // type - that is in the derived class and the two classes differ on
         // return type.
-
         assert(endpoint->getProtocol() == IPPROTO_UDP);
         const UDPEndpoint* udp_endpoint =
             static_cast<const UDPEndpoint*>(endpoint);
+
+        // ... and send the message.
         socket_.async_send_to(asio::buffer(data, length),
             udp_endpoint->getASIOEndpoint(), callback);
     } else {
@@ -229,14 +247,12 @@ UDPSocket<C>::asyncSend(const void* data, size_t length,
     }
 }
 
-// Receive a message. Note that the "cumulative" argument is ignored - every UDP
-// receive is put into the buffer beginning at the start - there is no concept
-// receiving a subsequent part of a message.  Same critera as before concerning
-// the need for the socket to be open.
+// Receive a message.   Should never do this if the socket is not open, so throw
+// an exception if this is the case.
 
 template <typename C> void
-UDPSocket<C>::asyncReceive(void* data, size_t length, size_t,
-    IOEndpoint* endpoint, C& callback)
+UDPSocket<C>::asyncReceive(void* data, size_t length, size_t offset,
+                           IOEndpoint* endpoint, C& callback)
 {
     if (isopen_) {
 
@@ -244,7 +260,15 @@ UDPSocket<C>::asyncReceive(void* data, size_t length, size_t,
         assert(endpoint->getProtocol() == IPPROTO_UDP);
         UDPEndpoint* udp_endpoint = static_cast<UDPEndpoint*>(endpoint);
 
-        socket_.async_receive_from(asio::buffer(data, length),
+        // Ensure we can write into the buffer
+        if (offset >= length) {
+            isc_throw(BufferOverflow, "attempt to read into area beyond end of "
+                                      "UDP receive buffer");
+        }
+        void* buffer_start = static_cast<void*>(static_cast<uint8_t*>(data) + offset);
+
+        // Issue the read
+        socket_.async_receive_from(asio::buffer(buffer_start, length - offset),
             udp_endpoint->getASIOEndpoint(), callback);
     } else {
         isc_throw(SocketNotOpen,
@@ -252,7 +276,29 @@ UDPSocket<C>::asyncReceive(void* data, size_t length, size_t,
     }
 }
 
+// Receive complete.  Just copy the data across to the output buffer and
+// update arguments as appropriate.
+
+template <typename C> bool
+UDPSocket<C>::processReceivedData(const void* staging, size_t length,
+                                  size_t& cumulative, size_t& offset,
+                                  size_t& expected,
+                                  isc::dns::OutputBufferPtr& outbuff)
+{
+    // Set return values to what we should expect.
+    cumulative = length;
+    expected = length;
+    offset = 0;
+
+    // Copy data across
+    outbuff->writeData(staging, length);
+
+    // ... and mark that we have everything.
+    return (true);
+}
+
 // Cancel I/O on the socket.  No-op if the socket is not open.
+
 template <typename C> void
 UDPSocket<C>::cancel() {
     if (isopen_) {

src/lib/cache/message_cache.cc

@@ -46,8 +46,16 @@ MessageCache::lookup(const isc::dns::Name& qname,
     HashKey entry_key = HashKey(entry_name, RRClass(message_class_));
     MessageEntryPtr msg_entry = message_table_.get(entry_key);
     if(msg_entry) {
-        message_lru_.touch(msg_entry);
-        return (msg_entry->genMessage(time(NULL), response));
+        // Check whether the message entry has expired.
+       if (msg_entry->getExpireTime() > time(NULL)) {
+            message_lru_.touch(msg_entry);
+            return (msg_entry->genMessage(time(NULL), response));
+        } else {
+            // message entry expires, remove it from hash table and lru list.
+            message_table_.remove(entry_key);
+            message_lru_.remove(msg_entry);
+            return (false);
+       }
     }
 
     return (false);

src/lib/cache/message_entry.h

@@ -87,6 +87,12 @@ public:
         return (*hash_key_ptr_);
     }
 
+    /// \brief Get expire time of the message entry.
+    /// \return return the expire time of message entry.
+    time_t getExpireTime() const {
+        return (expire_time_);
+    }
+
     /// \short Protected memebers, so they can be accessed by tests.
     //@{
 protected:

src/lib/cache/rrset_cache.cc

@@ -42,45 +42,41 @@ RRsetCache::lookup(const isc::dns::Name& qname,
 {
     const string entry_name = genCacheEntryName(qname, qtype);
     RRsetEntryPtr entry_ptr = rrset_table_.get(HashKey(entry_name, RRClass(class_)));
-
-    //If the rrset entry has expired, return NULL.
-    if(entry_ptr && (time(NULL) > entry_ptr->getExpireTime())) {
-        return (RRsetEntryPtr());
+    if (entry_ptr) {
+        if (entry_ptr->getExpireTime() > time(NULL)) {
+            // Only touch the non-expired rrset entries
+            rrset_lru_.touch(entry_ptr);
+            return (entry_ptr);
+        } else {
+            // the rrset entry has expired, so just remove it from
+            // hash table and lru list.
+            rrset_table_.remove(entry_ptr->hashKey());
+            rrset_lru_.remove(entry_ptr);
+        }
     }
-    return (entry_ptr);
+
+    return (RRsetEntryPtr());
 }
 
 RRsetEntryPtr
 RRsetCache::update(const isc::dns::RRset& rrset, const RRsetTrustLevel& level) {
     // TODO: If the RRset is an NS, we should update the NSAS as well
-    
     // lookup first
     RRsetEntryPtr entry_ptr = lookup(rrset.getName(), rrset.getType());
-    if(!entry_ptr) {
-        // rrset entry doesn't exist, create one rrset entry for the rrset
-        // and add it directly.
-        entry_ptr.reset(new RRsetEntry(rrset, level));
-        // Replace the expired rrset entry if it exists.
-        rrset_table_.add(entry_ptr, entry_ptr->hashKey(), true);
-        //TODO , lru list touch.
-        return (entry_ptr);
-    } else {
-        // there is one rrset entry in the cache, need to check whether
-        // the new rrset is more authoritative.
+    if (entry_ptr) {
         if (entry_ptr->getTrustLevel() > level) {
-            // existed rrset entry is more authoritative, do nothing,
-            // just return it.
-            //TODO, lru list touch
+            // existed rrset entry is more authoritative, just return it
             return (entry_ptr);
         } else {
-            HashKey key = entry_ptr->hashKey();
-            entry_ptr.reset(new RRsetEntry(rrset, level));
-            //TODO, lru list touch.
-            // Replace the expired rrset entry if it exists.
-            rrset_table_.add(entry_ptr, entry_ptr->hashKey(), true);
-            return (entry_ptr);
+            // Remove the old rrset entry from the lru list.
+            rrset_lru_.remove(entry_ptr);
         }
     }
+
+    entry_ptr.reset(new RRsetEntry(rrset, level));
+    rrset_table_.add(entry_ptr, entry_ptr->hashKey(), true);
+    rrset_lru_.add(entry_ptr);
+    return (entry_ptr);
 }
 
 #if 0

src/lib/cache/tests/Makefile.am

@@ -53,6 +53,7 @@ endif
 run_unittests_LDADD += $(top_builddir)/src/lib/cache/libcache.la
 run_unittests_LDADD += $(top_builddir)/src/lib/nsas/libnsas.la
 run_unittests_LDADD += $(top_builddir)/src/lib/dns/libdns++.la
+run_unittests_LDADD += $(top_builddir)/src/lib/asiolink/libasiolink.la
 run_unittests_LDADD += $(top_builddir)/src/lib/exceptions/libexceptions.la
 endif
 
@@ -66,3 +67,4 @@ EXTRA_DIST += testdata/message_fromWire5
 EXTRA_DIST += testdata/message_fromWire6
 EXTRA_DIST += testdata/message_fromWire7
 EXTRA_DIST += testdata/message_fromWire8
+EXTRA_DIST += testdata/message_fromWire9

src/lib/cache/tests/message_cache_unittest.cc

@@ -70,8 +70,8 @@ public:
     {
         uint16_t class_ = RRClass::IN().getCode();
         rrset_cache_.reset(new DerivedRRsetCache(RRSET_CACHE_DEFAULT_SIZE, class_));
-        message_cache_.reset(new DerivedMessageCache(rrset_cache_,
-                                          MESSAGE_CACHE_DEFAULT_SIZE, class_ ));
+        // Set the message cache size to 1, make it easy for unittest.
+        message_cache_.reset(new DerivedMessageCache(rrset_cache_, 1, class_ ));
     }
 
 protected:
@@ -81,9 +81,19 @@ protected:
     Message message_render;
 };
 
+void
+updateMessageCache(const char* message_file,
+                   boost::shared_ptr<DerivedMessageCache> cache)
+{
+    Message msg(Message::PARSE);
+    messageFromFile(msg, message_file);
+    cache->update(msg);
+}
+
 TEST_F(MessageCacheTest, testLookup) {
     messageFromFile(message_parse, "message_fromWire1");
     EXPECT_TRUE(message_cache_->update(message_parse));
+
     Name qname("test.example.com.");
     EXPECT_TRUE(message_cache_->lookup(qname, RRType::A(), message_render));
     EXPECT_EQ(message_cache_->messages_count(), 1);
@@ -96,10 +106,19 @@ TEST_F(MessageCacheTest, testLookup) {
     Name qname1("test.example.net.");
     EXPECT_TRUE(message_cache_->lookup(qname1, RRType::A(), message_render));
 
-    // Test looking up message which has expired rrsets.
-    // Remove one
+    // Test looking up message which has expired rrset or some rrset
+    // has been removed from the rrset cache.
     rrset_cache_->removeRRsetEntry(qname1, RRType::A());
     EXPECT_FALSE(message_cache_->lookup(qname1, RRType::A(), message_render));
+
+    // Update one message entry which has expired to message cache.
+    updateMessageCache("message_fromWire9", message_cache_);
+    EXPECT_EQ(message_cache_->messages_count(), 3);
+    // The message entry has been added, but can't be looked up, since
+    // it has expired and is removed automatically when being looked up.
+    Name qname_org("test.example.org.");
+    EXPECT_FALSE(message_cache_->lookup(qname_org, RRType::A(), message_render));
+    EXPECT_EQ(message_cache_->messages_count(), 2);
 }
 
 TEST_F(MessageCacheTest, testUpdate) {
@@ -118,5 +137,22 @@ TEST_F(MessageCacheTest, testUpdate) {
     EXPECT_TRUE(new_msg_render.getHeaderFlag(Message::HEADERFLAG_AA));
 }
 
+TEST_F(MessageCacheTest, testCacheLruBehavior) {
+    // qname = "test.example.com.", qtype = A
+    updateMessageCache("message_fromWire1", message_cache_);
+    // qname = "test.example.net.", qtype = A
+    updateMessageCache("message_fromWire2", message_cache_);
+    // qname = "example.com.", qtype = SOA
+    updateMessageCache("message_fromWire4", message_cache_);
+
+    Name qname_net("test.example.net.");
+    EXPECT_TRUE(message_cache_->lookup(qname_net, RRType::A(), message_render));
+
+    // qname = "a.example.com.", qtype = A
+    updateMessageCache("message_fromWire5", message_cache_);
+    Name qname_com("test.example.com.");
+    EXPECT_FALSE(message_cache_->lookup(qname_com, RRType::A(), message_render));
+}
+
 }   // namespace
 

src/lib/cache/tests/rrset_cache_unittest.cc

@@ -34,50 +34,94 @@ namespace {
 class RRsetCacheTest : public testing::Test {
 protected:
     RRsetCacheTest():
-        cache(RRSET_CACHE_DEFAULT_SIZE, RRClass::IN().getCode()),
-        name("example.com"),
-        rrset1(name, RRClass::IN(), RRType::A(), RRTTL(20)),
-        rrset2(name, RRClass::IN(), RRType::A(), RRTTL(10)),
-        rrset_entry1(rrset1, RRSET_TRUST_ADDITIONAL_AA),
-        rrset_entry2(rrset2, RRSET_TRUST_PRIM_ZONE_NONGLUE)
+        cache_(1, RRClass::IN().getCode()),
+        name_("example.com"),
+        rrset1_(name_, RRClass::IN(), RRType::A(), RRTTL(20)),
+        rrset2_(name_, RRClass::IN(), RRType::A(), RRTTL(10)),
+        rrset_entry1_(rrset1_, RRSET_TRUST_ADDITIONAL_AA),
+        rrset_entry2_(rrset2_, RRSET_TRUST_PRIM_ZONE_NONGLUE)
     {
     }
 
-    RRsetCache cache;
-    Name name;
-    RRset rrset1;
-    RRset rrset2;
-    RRsetEntry rrset_entry1;
-    RRsetEntry rrset_entry2;
+    RRsetCache cache_;
+    Name name_;
+    RRset rrset1_;
+    RRset rrset2_;
+    RRsetEntry rrset_entry1_;
+    RRsetEntry rrset_entry2_;
 };
 
+void
+updateRRsetCache(RRsetCache& cache, Name& rrset_name,
+                 uint32_t ttl = 20,
+                 RRsetTrustLevel level = RRSET_TRUST_ADDITIONAL_AA)
+{
+    RRset rrset(rrset_name, RRClass::IN(), RRType::A(), RRTTL(ttl));
+    cache.update(rrset, level);
+}
+
 TEST_F(RRsetCacheTest, lookup) {
     const RRType& type = RRType::A();
-    EXPECT_TRUE(cache.lookup(name, type) == NULL);
-
-    cache.update(rrset1, rrset_entry1.getTrustLevel());
-    RRsetEntryPtr rrset_entry_ptr = cache.lookup(name, type);
-    EXPECT_EQ(rrset_entry_ptr->getTrustLevel(), rrset_entry1.getTrustLevel());
-    EXPECT_EQ(rrset_entry_ptr->getRRset()->getName(), rrset_entry1.getRRset()->getName());
-    EXPECT_EQ(rrset_entry_ptr->getRRset()->getType(), rrset_entry1.getRRset()->getType());
-    EXPECT_EQ(rrset_entry_ptr->getRRset()->getClass(), rrset_entry1.getRRset()->getClass());
+    EXPECT_TRUE(cache_.lookup(name_, type) == NULL);
+
+    cache_.update(rrset1_, rrset_entry1_.getTrustLevel());
+    RRsetEntryPtr rrset_entry_ptr = cache_.lookup(name_, type);
+    EXPECT_EQ(rrset_entry_ptr->getTrustLevel(), rrset_entry1_.getTrustLevel());
+    EXPECT_EQ(rrset_entry_ptr->getRRset()->getName(), rrset_entry1_.getRRset()->getName());
+    EXPECT_EQ(rrset_entry_ptr->getRRset()->getType(), rrset_entry1_.getRRset()->getType());
+    EXPECT_EQ(rrset_entry_ptr->getRRset()->getClass(), rrset_entry1_.getRRset()->getClass());
+
+    // Check whether the expired rrset entry will be removed automatically
+    // when looking up.
+    Name name_test("test.example.com.");
+    updateRRsetCache(cache_, name_test, 0); // Add a rrset with TTL 0 to cache.
+    EXPECT_FALSE(cache_.lookup(name_test, RRType::A()));
 }
 
 TEST_F(RRsetCacheTest, update) {
     const RRType& type = RRType::A();
 
-    cache.update(rrset1, rrset_entry1.getTrustLevel());
-    RRsetEntryPtr rrset_entry_ptr = cache.lookup(name, type);
-    EXPECT_EQ(rrset_entry_ptr->getTrustLevel(), rrset_entry1.getTrustLevel());
+    cache_.update(rrset1_, rrset_entry1_.getTrustLevel());
+    RRsetEntryPtr rrset_entry_ptr = cache_.lookup(name_, type);
+    EXPECT_EQ(rrset_entry_ptr->getTrustLevel(), rrset_entry1_.getTrustLevel());
 
-    cache.update(rrset2, rrset_entry2.getTrustLevel());
-    rrset_entry_ptr = cache.lookup(name, type);
+    cache_.update(rrset2_, rrset_entry2_.getTrustLevel());
+    rrset_entry_ptr = cache_.lookup(name_, type);
     // The trust level should be updated
-    EXPECT_EQ(rrset_entry_ptr->getTrustLevel(), rrset_entry2.getTrustLevel());
+    EXPECT_EQ(rrset_entry_ptr->getTrustLevel(), rrset_entry2_.getTrustLevel());
 
-    cache.update(rrset1, rrset_entry1.getTrustLevel());
+    cache_.update(rrset1_, rrset_entry1_.getTrustLevel());
     // The trust level should not be updated
-    EXPECT_EQ(rrset_entry_ptr->getTrustLevel(), rrset_entry2.getTrustLevel());
+    EXPECT_EQ(rrset_entry_ptr->getTrustLevel(), rrset_entry2_.getTrustLevel());
+}
+
+// Test whether the lru list in rrset cache works as expected.
+TEST_F(RRsetCacheTest, cacheLruBehavior) {
+    Name name1("1.example.com.");
+    Name name2("2.example.com.");
+    Name name3("3.example.com.");
+    Name name4("4.example.com.");
+
+    updateRRsetCache(cache_, name1);
+    updateRRsetCache(cache_, name2);
+    updateRRsetCache(cache_, name3);
+
+    EXPECT_TRUE(cache_.lookup(name1, RRType::A()));
+
+    // Now update the fourth rrset, rrset with name "2.example.com."
+    // should has been removed from cache.
+    updateRRsetCache(cache_, name4);
+    EXPECT_FALSE(cache_.lookup(name2, RRType::A()));
+
+    // Test Update rrset with higher trust level
+    updateRRsetCache(cache_, name1, RRSET_TRUST_PRIM_GLUE);
+    // Test update rrset with lower trust level.
+    updateRRsetCache(cache_, name3, RRSET_TRUST_ADDITIONAL_NONAA);
+
+    // When add rrset with name2, rrset with name4
+    // has been removed from the cache.
+    updateRRsetCache(cache_, name2);
+    EXPECT_FALSE(cache_.lookup(name4, RRType::A()));
 }
 
 }

src/lib/cache/tests/testdata/message_fromWire9

@@ -0,0 +1,25 @@
+#
+# The TTL for a record in answer section is 0, so it
+# will expire immediately after being cached.
+#
+# A simple DNS response message
+# ID = 0x1035
+# QR=1 (response), Opcode=0, AA=1, RD=1 (other fields are 0)
+# QDCOUNT=1, ANCOUNT=2, other COUNTS=0
+# Question: test.example.org. IN A
+# Answer:
+#  test.example.org. 0000 IN A 192.0.2.1
+#  test.example.org. 7200 IN A 192.0.2.2
+#
+1035 8500
+0001 0002 0000 0000
+#(4) t  e  s  t (7) e  x  a  m  p  l  e (3) o  r  g  .
+ 04 74 65 73 74 07 65 78 61 6d 70 6c 65 03 6f 72 67 00
+0001 0001
+# same name, fully compressed
+c0 0c
+# TTL=3600, A, IN, RDLENGTH=4, RDATA
+0001 0001 00000000 0004 c0 00 02 01
+# mostly same, with the slight difference in RDATA and TTL
+c0 0c
+0001 0001 00001c20 0004 c0 00 02 02

src/lib/datasrc/data_source.cc

@@ -189,6 +189,19 @@ checkCache(QueryTask& task, RRsetList& target) {
                 rrsets.addRRset(rrset);
                 target.append(rrsets);
             }
+
+            // Reset the referral flag and treat CNAME as "not found".
+            // This emulates the behavior of the sqlite3 data source.
+            // XXX: this is not ideal in that the responsibility for handling
+            // operation specific cases is spread over various classes at
+            // different abstraction levels.  For longer terms we should
+            // revisit the whole datasource/query design, and clarify this
+            // point better.
+            flags &= ~DataSrc::REFERRAL;
+            if ((flags & DataSrc::CNAME_FOUND) != 0) {
+                flags &= ~DataSrc::CNAME_FOUND;
+                flags |= DataSrc::TYPE_NOT_FOUND;
+            }
             task.flags = flags;
             return (true);
         }

src/lib/datasrc/rbtree.h

@@ -533,12 +533,9 @@ private:
 
 private:
     // The max label count for one domain name is Name::MAX_LABELS (128).
-    // Since each node in rbtree stores at least one label, and the root
-    // name always shares the same level with some others (which means
-    // all top level nodes except the one for the root name contain at least
-    // two labels), the possible maximum level is MAX_LABELS - 1.
-    // It's also the possible maximum nodes stored in a chain.
-    const static int RBT_MAX_LEVEL = isc::dns::Name::MAX_LABELS - 1;
+    // Since each node in rbtree stores at least one label, it's also equal
+    // to the possible maximum level.
+    const static int RBT_MAX_LEVEL = isc::dns::Name::MAX_LABELS;
 
     int node_count_;
     const RBNode<T>* nodes_[RBT_MAX_LEVEL];
@@ -999,8 +996,15 @@ RBTree<T>::find(const isc::dns::Name& target_name,
             const int common_label_count =
                 node_path.last_comparison_.getCommonLabels();
             // If the common label count is 1, there is no common label between
-            // the two names, except the trailing "dot".
-            if (common_label_count == 1) {
+            // the two names, except the trailing "dot".  In this case the two
+            // sequences of labels have essentially no hierarchical
+            // relationship in terms of matching, so we should continue the
+            // binary search.  One important exception is when the node
+            // represents the root name ("."), in which case the comparison
+            // result must indeed be considered subdomain matching. (We use
+            // getLength() to check if the name is root, which is an equivalent
+            // but cheaper way).
+            if (common_label_count == 1 && node->name_.getLength() != 1) {
                 node = (node_path.last_comparison_.getOrder() < 0) ?
                     node->left_ : node->right_;
             } else if (relation == isc::dns::NameComparisonResult::SUBDOMAIN) {
@@ -1093,7 +1097,8 @@ RBTree<T>::insert(const isc::dns::Name& target_name, RBNode<T>** new_node) {
             return (ALREADYEXISTS);
         } else {
             const int common_label_count = compare_result.getCommonLabels();
-            if (common_label_count == 1) {
+            // Note: see find() for the check of getLength().
+            if (common_label_count == 1 && current->name_.getLength() != 1) {
                 parent = current;
                 order = compare_result.getOrder();
                 current = order < 0 ? current->left_ : current->right_;

src/lib/datasrc/tests/Makefile.am

@@ -31,6 +31,7 @@ run_unittests_CPPFLAGS = $(AM_CPPFLAGS) $(GTEST_INCLUDES)
 run_unittests_LDFLAGS = $(AM_LDFLAGS) $(GTEST_LDFLAGS)
 run_unittests_LDADD = $(GTEST_LDADD)
 run_unittests_LDADD += $(SQLITE_LIBS)
+run_unittests_LDADD += $(top_builddir)/src/lib/testutils/libtestutils.la
 run_unittests_LDADD += $(top_builddir)/src/lib/datasrc/libdatasrc.la
 run_unittests_LDADD += $(top_builddir)/src/lib/dns/libdns++.la
 run_unittests_LDADD += $(top_builddir)/src/lib/cc/libcc.la

src/lib/datasrc/tests/datasrc_unittest.cc

@@ -38,6 +38,7 @@
 #include <datasrc/sqlite3_datasrc.h>
 #include <datasrc/static_datasrc.h>
 
+#include <testutils/dnsmessage_test.h>
 #include <dns/tests/unittest_util.h>
 #include <datasrc/tests/test_datasrc.h>
 
@@ -47,6 +48,7 @@ using namespace isc::dns;
 using namespace isc::dns::rdata;
 using namespace isc::datasrc;
 using namespace isc::data;
+using namespace isc::testutils;
 
 namespace {
 ConstElementPtr SQLITE_DBFILE_EXAMPLE = Element::fromJSON(
@@ -54,7 +56,9 @@ ConstElementPtr SQLITE_DBFILE_EXAMPLE = Element::fromJSON(
 
 class DataSrcTest : public ::testing::Test {
 protected:
-    DataSrcTest() : obuffer(0), renderer(obuffer), msg(Message::PARSE) {
+    DataSrcTest() : obuffer(0), renderer(obuffer), msg(Message::PARSE),
+                    opcodeval(Opcode::QUERY().getCode()), qid(0)
+    {
         DataSrcPtr sql3_source = DataSrcPtr(new Sqlite3DataSrc); 
         sql3_source->init(SQLITE_DBFILE_EXAMPLE);
         DataSrcPtr test_source = DataSrcPtr(new TestDataSrc);
@@ -73,6 +77,8 @@ protected:
     OutputBuffer obuffer;
     MessageRenderer renderer;
     Message msg;
+    const uint16_t opcodeval;
+    qid_t qid;
 };
 
 void
@@ -91,29 +97,16 @@ DataSrcTest::createAndProcessQuery(const Name& qname, const RRClass& qclass,
     msg.setOpcode(Opcode::QUERY());
     msg.addQuestion(Question(qname, qclass, qtype));
     msg.setHeaderFlag(Message::HEADERFLAG_RD);
+    qid = msg.getQid();
     performQuery(meta_source, cache, msg);
 }
 
 void
-headerCheck(const Message& message, const Rcode& rcode, const bool qrflag,
-            const bool aaflag, const bool rdflag, const unsigned int ancount,
-            const unsigned int nscount, const unsigned int arcount)
-{
-    EXPECT_EQ(rcode, message.getRcode());
-    EXPECT_EQ(qrflag, message.getHeaderFlag(Message::HEADERFLAG_QR));
-    EXPECT_EQ(aaflag, message.getHeaderFlag(Message::HEADERFLAG_AA));
-    EXPECT_EQ(rdflag, message.getHeaderFlag(Message::HEADERFLAG_RD));
-
-    EXPECT_EQ(ancount, message.getRRCount(Message::SECTION_ANSWER));
-    EXPECT_EQ(nscount, message.getRRCount(Message::SECTION_AUTHORITY));
-    EXPECT_EQ(arcount, message.getRRCount(Message::SECTION_ADDITIONAL));
-}
-
-void
 DataSrcTest::QueryCommon(const RRClass& qclass) {
     createAndProcessQuery(Name("www.example.com"), qclass, RRType::A());
 
-    headerCheck(msg, Rcode::NOERROR(), true, true, true, 2, 4, 6);
+    headerCheck(msg, qid, Rcode::NOERROR(), opcodeval,
+                QR_FLAG | AA_FLAG | RD_FLAG, 1, 2, 4, 6);
 
     RRsetIterator rit = msg.beginSection(Message::SECTION_ANSWER);
     RRsetPtr rrset = *rit;
@@ -163,12 +156,8 @@ TEST_F(DataSrcTest, Query) {
 // should be the same as "NxZone".
 TEST_F(DataSrcTest, QueryClassMismatch) {
     createAndProcessQuery(Name("www.example.com"), RRClass::CH(), RRType::A());
-    headerCheck(msg, Rcode::REFUSED(), true, false, true, 0, 0, 0);
-
-    EXPECT_EQ(Rcode::REFUSED(), msg.getRcode());
-    EXPECT_TRUE(msg.getHeaderFlag(Message::HEADERFLAG_QR));
-    EXPECT_FALSE(msg.getHeaderFlag(Message::HEADERFLAG_AA));
-    EXPECT_TRUE(msg.getHeaderFlag(Message::HEADERFLAG_RD));
+    headerCheck(msg, qid, Rcode::REFUSED(), opcodeval, QR_FLAG | RD_FLAG,
+                1, 0, 0, 0);
 }
 
 // Query class of any should match the first data source.
@@ -179,7 +168,8 @@ TEST_F(DataSrcTest, QueryClassAny) {
 TEST_F(DataSrcTest, NSQuery) {
     createAndProcessQuery(Name("example.com"), RRClass::IN(),
                           RRType::NS());
-    headerCheck(msg, Rcode::NOERROR(), true, true, true, 4, 0, 6);
+    headerCheck(msg, qid, Rcode::NOERROR(), opcodeval,
+                QR_FLAG | AA_FLAG | RD_FLAG, 1, 4, 0, 6);
 
     RRsetIterator rit = msg.beginSection(Message::SECTION_ANSWER);
     RRsetPtr rrset = *rit;
@@ -201,7 +191,8 @@ TEST_F(DataSrcTest, NSQuery) {
 TEST_F(DataSrcTest, DuplicateQuery) {
     createAndProcessQuery(Name("example.com"), RRClass::IN(),
                           RRType::NS());
-    headerCheck(msg, Rcode::NOERROR(), true, true, true, 4, 0, 6);
+    headerCheck(msg, qid, Rcode::NOERROR(), opcodeval,
+                QR_FLAG | AA_FLAG | RD_FLAG, 1, 4, 0, 6);
 
     RRsetIterator rit = msg.beginSection(Message::SECTION_ANSWER);
     RRsetPtr rrset = *rit;
@@ -221,7 +212,8 @@ TEST_F(DataSrcTest, DuplicateQuery) {
     msg.clear(Message::PARSE);
     createAndProcessQuery(Name("example.com"), RRClass::IN(),
                           RRType::NS());
-    headerCheck(msg, Rcode::NOERROR(), true, true, true, 4, 0, 6);
+    headerCheck(msg, qid, Rcode::NOERROR(), opcodeval,
+                QR_FLAG | AA_FLAG | RD_FLAG, 1, 4, 0, 6);
 
     rit = msg.beginSection(Message::SECTION_ANSWER);
     rrset = *rit;
@@ -242,7 +234,8 @@ TEST_F(DataSrcTest, DuplicateQuery) {
 TEST_F(DataSrcTest, DNSKEYQuery) {
     createAndProcessQuery(Name("example.com"), RRClass::IN(),
                           RRType::DNSKEY());
-    headerCheck(msg, Rcode::NOERROR(), true, true, true, 4, 4, 6);
+    headerCheck(msg, qid, Rcode::NOERROR(), opcodeval,
+                QR_FLAG | AA_FLAG | RD_FLAG, 1, 4, 4, 6);
 
     RRsetIterator rit = msg.beginSection(Message::SECTION_ANSWER);
     RRsetPtr rrset = *rit;
@@ -257,7 +250,8 @@ TEST_F(DataSrcTest, DNSKEYQuery) {
 TEST_F(DataSrcTest, DNSKEYDuplicateQuery) {
     createAndProcessQuery(Name("example.com"), RRClass::IN(),
                           RRType::DNSKEY());
-    headerCheck(msg, Rcode::NOERROR(), true, true, true, 4, 4, 6);
+    headerCheck(msg, qid, Rcode::NOERROR(), opcodeval,
+                QR_FLAG | AA_FLAG | RD_FLAG, 1, 4, 4, 6);
 
     RRsetIterator rit = msg.beginSection(Message::SECTION_ANSWER);
     RRsetPtr rrset = *rit;
@@ -279,7 +273,8 @@ TEST_F(DataSrcTest, NxRRset) {
     createAndProcessQuery(Name("example.com"), RRClass::IN(),
                           RRType::PTR());
 
-    headerCheck(msg, Rcode::NOERROR(), true, true, true, 0, 4, 0);
+    headerCheck(msg, qid, Rcode::NOERROR(), opcodeval,
+                QR_FLAG | AA_FLAG | RD_FLAG, 1, 0, 4, 0);
 
     RRsetIterator rit = msg.beginSection(Message::SECTION_AUTHORITY);
     RRsetPtr rrset = *rit;
@@ -291,7 +286,8 @@ TEST_F(DataSrcTest, Nxdomain) {
     createAndProcessQuery(Name("glork.example.com"), RRClass::IN(),
                           RRType::A());
 
-    headerCheck(msg, Rcode::NXDOMAIN(), true, true, true, 0, 6, 0);
+    headerCheck(msg, qid, Rcode::NXDOMAIN(), opcodeval,
+                QR_FLAG | AA_FLAG | RD_FLAG, 1, 0, 6, 0);
 
     RRsetIterator rit = msg.beginSection(Message::SECTION_AUTHORITY);
     RRsetPtr rrset = *rit;
@@ -301,11 +297,46 @@ TEST_F(DataSrcTest, Nxdomain) {
     // XXX: check for other authority section answers
 }
 
+TEST_F(DataSrcTest, NxdomainAfterSOAQuery) {
+    // There was a bug where once SOA RR is stored in the hot spot cache
+    // subsequent negative search fails due to "missing SOA".  This test
+    // checks that situation.
+
+    // First, run the scenario with disabling the cache.
+    cache.setEnabled(false);
+    createAndProcessQuery(Name("example.com"), RRClass::IN(),
+                          RRType::SOA());
+    msg.clear(Message::PARSE);
+    createAndProcessQuery(Name("notexistent.example.com"), RRClass::IN(),
+                          RRType::A());
+    {
+        SCOPED_TRACE("NXDOMAIN after SOA, without hot spot cache");
+        headerCheck(msg, qid, Rcode::NXDOMAIN(), opcodeval,
+                    QR_FLAG | AA_FLAG | RD_FLAG, 1, 0, 6, 0);
+    }
+
+    // Then enable the cache and perform the same queries.  This should
+    // produce the same result.
+    cache.setEnabled(true);
+    msg.clear(Message::PARSE);
+    createAndProcessQuery(Name("example.com"), RRClass::IN(),
+                          RRType::SOA());
+    msg.clear(Message::PARSE);
+    createAndProcessQuery(Name("notexistent.example.com"), RRClass::IN(),
+                        RRType::A());
+    {
+        SCOPED_TRACE("NXDOMAIN after SOA, without hot spot cache");
+        headerCheck(msg, qid, Rcode::NXDOMAIN(), opcodeval,
+                    QR_FLAG | AA_FLAG | RD_FLAG, 1, 0, 6, 0);
+    }
+}
+
 TEST_F(DataSrcTest, NxZone) {
     createAndProcessQuery(Name("spork.example"), RRClass::IN(),
                           RRType::A());
 
-    headerCheck(msg, Rcode::REFUSED(), true, false, true, 0, 0, 0);
+    headerCheck(msg, qid, Rcode::REFUSED(), opcodeval,
+                QR_FLAG | RD_FLAG, 1, 0, 0, 0);
 
     EXPECT_EQ(Rcode::REFUSED(), msg.getRcode());
     EXPECT_TRUE(msg.getHeaderFlag(Message::HEADERFLAG_QR));
@@ -317,7 +348,8 @@ TEST_F(DataSrcTest, Wildcard) {
     createAndProcessQuery(Name("www.wild.example.com"), RRClass::IN(),
                           RRType::A());
 
-    headerCheck(msg, Rcode::NOERROR(), true, true, true, 2, 6, 6);
+    headerCheck(msg, qid, Rcode::NOERROR(), opcodeval,
+                QR_FLAG | AA_FLAG | RD_FLAG, 1, 2, 6, 6);
 
     RRsetIterator rit = msg.beginSection(Message::SECTION_ANSWER);
     RRsetPtr rrset = *rit;
@@ -369,7 +401,8 @@ TEST_F(DataSrcTest, WildcardNodata) {
     // returns NOERROR
     createAndProcessQuery(Name("www.wild.example.com"), RRClass::IN(),
                           RRType::AAAA());
-    headerCheck(msg, Rcode::NOERROR(), true, true, true, 0, 2, 0);
+    headerCheck(msg, qid, Rcode::NOERROR(), opcodeval,
+                QR_FLAG | AA_FLAG | RD_FLAG, 1, 0, 2, 0);
 }
 
 TEST_F(DataSrcTest, DISABLED_WildcardAgainstMultiLabel) {
@@ -377,7 +410,8 @@ TEST_F(DataSrcTest, DISABLED_WildcardAgainstMultiLabel) {
     // a single label), and it should result in NXDOMAIN.
     createAndProcessQuery(Name("www.xxx.wild.example.com"), RRClass::IN(),
                           RRType::A());
-    headerCheck(msg, Rcode::NXDOMAIN(), true, true, true, 0, 1, 0);
+    headerCheck(msg, qid, Rcode::NXDOMAIN(), opcodeval,
+                QR_FLAG | AA_FLAG | RD_FLAG, 1, 0, 1, 0);
 }
 
 TEST_F(DataSrcTest, WildcardCname) {
@@ -386,7 +420,8 @@ TEST_F(DataSrcTest, WildcardCname) {
     createAndProcessQuery(Name("www.wild2.example.com"), RRClass::IN(),
                           RRType::A());
 
-    headerCheck(msg, Rcode::NOERROR(), true, true, true, 4, 6, 6);
+    headerCheck(msg, qid, Rcode::NOERROR(), opcodeval,
+                QR_FLAG | AA_FLAG | RD_FLAG, 1, 4, 6, 6);
 
     RRsetIterator rit = msg.beginSection(Message::SECTION_ANSWER);
     RRsetPtr rrset = *rit;
@@ -450,7 +485,8 @@ TEST_F(DataSrcTest, WildcardCnameNodata) {
     // data of this type.
     createAndProcessQuery(Name("www.wild2.example.com"), RRClass::IN(),
                           RRType::AAAA());
-    headerCheck(msg, Rcode::NOERROR(), true, true, true, 2, 4, 0);
+    headerCheck(msg, qid, Rcode::NOERROR(), opcodeval,
+                QR_FLAG | AA_FLAG | RD_FLAG, 1, 2, 4, 0);
 
     RRsetIterator rit = msg.beginSection(Message::SECTION_ANSWER);
     RRsetPtr rrset = *rit;
@@ -481,7 +517,8 @@ TEST_F(DataSrcTest, WildcardCnameNxdomain) {
     // A wildcard containing a CNAME whose target does not exist
     createAndProcessQuery(Name("www.wild3.example.com"), RRClass::IN(),
                           RRType::A());
-    headerCheck(msg, Rcode::NOERROR(), true, true, true, 2, 6, 0);
+    headerCheck(msg, qid, Rcode::NOERROR(), opcodeval,
+                QR_FLAG | AA_FLAG | RD_FLAG, 1, 2, 6, 0);
 
     RRsetIterator rit = msg.beginSection(Message::SECTION_ANSWER);
     RRsetPtr rrset = *rit;
@@ -518,7 +555,8 @@ TEST_F(DataSrcTest, AuthDelegation) {
     createAndProcessQuery(Name("www.sql1.example.com"), RRClass::IN(),
                           RRType::A());
 
-    headerCheck(msg, Rcode::NOERROR(), true, true, true, 2, 4, 6);
+    headerCheck(msg, qid, Rcode::NOERROR(), opcodeval,
+                QR_FLAG | AA_FLAG | RD_FLAG, 1, 2, 4, 6);
 
     RRsetIterator rit = msg.beginSection(Message::SECTION_ANSWER);
     RRsetPtr rrset = *rit;
@@ -562,7 +600,8 @@ TEST_F(DataSrcTest, Dname) {
     createAndProcessQuery(Name("www.dname.example.com"), RRClass::IN(),
                           RRType::A());
 
-    headerCheck(msg, Rcode::NOERROR(), true, true, true, 5, 4, 6);
+    headerCheck(msg, qid, Rcode::NOERROR(), opcodeval,
+                QR_FLAG | AA_FLAG | RD_FLAG, 1, 5, 4, 6);
 
     RRsetIterator rit = msg.beginSection(Message::SECTION_ANSWER);
     RRsetPtr rrset = *rit;
@@ -610,14 +649,16 @@ TEST_F(DataSrcTest, DnameExact) {
     // confuse delegation processing.
     createAndProcessQuery(Name("dname2.foo.example.org"), RRClass::IN(),
                           RRType::A());
-    headerCheck(msg, Rcode::NOERROR(), true, true, true, 0, 1, 0);
+    headerCheck(msg, qid, Rcode::NOERROR(), opcodeval,
+                QR_FLAG | AA_FLAG | RD_FLAG, 1, 0, 1, 0);
 }
 
 TEST_F(DataSrcTest, Cname) {
     createAndProcessQuery(Name("foo.example.com"), RRClass::IN(),
                           RRType::A());
 
-    headerCheck(msg, Rcode::NOERROR(), true, true, true, 2, 0, 0);
+    headerCheck(msg, qid, Rcode::NOERROR(), opcodeval,
+                QR_FLAG | AA_FLAG | RD_FLAG, 1, 2, 0, 0);
 
     RRsetIterator rit = msg.beginSection(Message::SECTION_ANSWER);
     RRsetPtr rrset = *rit;
@@ -635,7 +676,8 @@ TEST_F(DataSrcTest, CnameInt) {
     createAndProcessQuery(Name("cname-int.example.com"), RRClass::IN(),
                           RRType::A());
 
-    headerCheck(msg, Rcode::NOERROR(), true, true, true, 4, 4, 6);
+    headerCheck(msg, qid, Rcode::NOERROR(), opcodeval,
+                QR_FLAG | AA_FLAG | RD_FLAG, 1, 4, 4, 6);
 
     RRsetIterator rit = msg.beginSection(Message::SECTION_ANSWER);
     RRsetPtr rrset = *rit;
@@ -661,7 +703,8 @@ TEST_F(DataSrcTest, CnameExt) {
     createAndProcessQuery(Name("cname-ext.example.com"), RRClass::IN(),
                           RRType::A());
 
-    headerCheck(msg, Rcode::NOERROR(), true, true, true, 4, 4, 6);
+    headerCheck(msg, qid, Rcode::NOERROR(), opcodeval,
+                QR_FLAG | AA_FLAG | RD_FLAG, 1, 4, 4, 6);
 
     RRsetIterator rit = msg.beginSection(Message::SECTION_ANSWER);
     RRsetPtr rrset = *rit;
@@ -685,7 +728,8 @@ TEST_F(DataSrcTest, Delegation) {
     createAndProcessQuery(Name("www.subzone.example.com"), RRClass::IN(),
                           RRType::A());
 
-    headerCheck(msg, Rcode::NOERROR(), true, false, true, 0, 5, 2);
+    headerCheck(msg, qid, Rcode::NOERROR(), opcodeval,
+                QR_FLAG | RD_FLAG, 1, 0, 5, 2);
 
     RRsetIterator rit = msg.beginSection(Message::SECTION_AUTHORITY);
     RRsetPtr rrset = *rit;
@@ -714,7 +758,8 @@ TEST_F(DataSrcTest, NSDelegation) {
     createAndProcessQuery(Name("subzone.example.com"), RRClass::IN(),
                           RRType::NS());
 
-    headerCheck(msg, Rcode::NOERROR(), true, false, true, 0, 5, 2);
+    headerCheck(msg, qid, Rcode::NOERROR(), opcodeval,
+                QR_FLAG | RD_FLAG, 1, 0, 5, 2);
 
     RRsetIterator rit = msg.beginSection(Message::SECTION_AUTHORITY);
     RRsetPtr rrset = *rit;
@@ -750,7 +795,8 @@ TEST_F(DataSrcTest, NSECZonecut) {
     createAndProcessQuery(Name("subzone.example.com"), RRClass::IN(),
                           RRType::NSEC());
 
-    headerCheck(msg, Rcode::NOERROR(), true, true, true, 2, 4, 6);
+    headerCheck(msg, qid, Rcode::NOERROR(), opcodeval,
+                QR_FLAG | AA_FLAG | RD_FLAG, 1, 2, 4, 6);
 
     RRsetIterator rit = msg.beginSection(Message::SECTION_ANSWER);
     RRsetPtr rrset = *rit;
@@ -778,7 +824,8 @@ TEST_F(DataSrcTest, DNAMEZonecut) {
     createAndProcessQuery(Name("subzone.example.com"), RRClass::IN(),
                           RRType::DNAME());
 
-    headerCheck(msg, Rcode::NOERROR(), true, false, true, 0, 5, 2);
+    headerCheck(msg, qid, Rcode::NOERROR(), opcodeval,
+                QR_FLAG | RD_FLAG, 1, 0, 5, 2);
     RRsetIterator rit = msg.beginSection(Message::SECTION_AUTHORITY);
     RRsetPtr rrset = *rit;
     EXPECT_EQ(Name("subzone.example.com."), rrset->getName());
@@ -806,7 +853,8 @@ TEST_F(DataSrcTest, DS) {
     createAndProcessQuery(Name("subzone.example.com"), RRClass::IN(),
                           RRType::DS());
 
-    headerCheck(msg, Rcode::NOERROR(), true, true, true, 3, 4, 6);
+    headerCheck(msg, qid, Rcode::NOERROR(), opcodeval,
+                QR_FLAG | AA_FLAG | RD_FLAG, 1, 3, 4, 6);
 
     RRsetIterator rit = msg.beginSection(Message::SECTION_ANSWER);
     RRsetPtr rrset = *rit;
@@ -847,7 +895,8 @@ TEST_F(DataSrcTest, NSECZonecutOfNonsecureZone) {
     createAndProcessQuery(Name("sub.example.org"), RRClass::IN(),
                           RRType::NSEC());
 
-    headerCheck(msg, Rcode::NOERROR(), true, false, true, 0, 1, 1);
+    headerCheck(msg, qid, Rcode::NOERROR(), opcodeval,
+                QR_FLAG | RD_FLAG, 1, 0, 1, 1);
 
     RRsetIterator rit = msg.beginSection(Message::SECTION_AUTHORITY);
     ConstRRsetPtr rrset = *rit;
@@ -879,7 +928,8 @@ TEST_F(DataSrcTest, NSECZonecutOfNonsecureZone) {
 TEST_F(DataSrcTest, RootDSQuery1) {
     EXPECT_NO_THROW(createAndProcessQuery(Name("."), RRClass::IN(),
                                           RRType::DS()));
-    headerCheck(msg, Rcode::REFUSED(), true, false, true, 0, 0, 0);
+    headerCheck(msg, qid, Rcode::REFUSED(), opcodeval,
+                QR_FLAG | RD_FLAG, 1, 0, 0, 0);
 }
 
 // The same, but when we have the root zone
@@ -898,7 +948,8 @@ TEST_F(DataSrcTest, RootDSQuery2) {
     // Make the query
     EXPECT_NO_THROW(performQuery(*sql3_source, cache, msg));
 
-    headerCheck(msg, Rcode::NOERROR(), true, true, true, 0, 1, 0);
+    headerCheck(msg, qid, Rcode::NOERROR(), opcodeval,
+                QR_FLAG | AA_FLAG | RD_FLAG, 1, 0, 1, 0);
 }
 
 TEST_F(DataSrcTest, DSQueryFromCache) {
@@ -916,7 +967,8 @@ TEST_F(DataSrcTest, DSQueryFromCache) {
 
     // returning refused is probably a bad behavior, but it's a different
     // issue -- see Trac Ticket #306.
-    headerCheck(msg, Rcode::REFUSED(), true, false, true, 0, 0, 0);
+    headerCheck(msg, qid, Rcode::REFUSED(), opcodeval,
+                QR_FLAG | RD_FLAG, 1, 0, 0, 0);
 }
 
 // Non-existent name in the "static" data source.  The purpose of this test
@@ -925,7 +977,8 @@ TEST_F(DataSrcTest, DSQueryFromCache) {
 TEST_F(DataSrcTest, StaticNxDomain) {
     createAndProcessQuery(Name("www.version.bind"), RRClass::CH(),
                           RRType::TXT());
-    headerCheck(msg, Rcode::NXDOMAIN(), true, true, true, 0, 1, 0);
+    headerCheck(msg, qid, Rcode::NXDOMAIN(), opcodeval,
+                QR_FLAG | AA_FLAG | RD_FLAG, 1, 0, 1, 0);
     RRsetIterator rit = msg.beginSection(Message::SECTION_AUTHORITY);
     RRsetPtr rrset = *rit;
     EXPECT_EQ(Name("version.bind"), rrset->getName());
@@ -973,6 +1026,15 @@ TEST_F(DataSrcTest, noSOAZone) {
                  DataSourceError);
 }
 
+TEST_F(DataSrcTest, apexCNAMEZone) {
+    // The query name doesn't exist in the best matching zone,
+    // and there's a CNAME at the apex (which is bogus), so query handling
+    // will fail due to missing SOA.
+    EXPECT_THROW(createAndProcessQuery(Name("notexist.apexcname.example"),
+                                       RRClass::IN(), RRType::A()),
+                 DataSourceError);
+}
+
 // currently fails
 TEST_F(DataSrcTest, DISABLED_synthesizedCnameTooLong) {
     // qname has the possible max length (255 octets).  it matches a DNAME,

src/lib/datasrc/tests/rbtree_unittest.cc

@@ -284,21 +284,21 @@ TEST_F(RBTreeTest, chainLevel) {
     EXPECT_EQ(1, chain.getLevelCount());
 
     /*
-     * Now creating a possibly deepest tree with MAX_LABELS - 1 levels.
+     * Now creating a possibly deepest tree with MAX_LABELS levels.
      * it should look like:
+     *           (.)
+     *            |
      *            a
-     *           /|
-     *         (.)a
      *            |
      *            a
      *            : (MAX_LABELS - 1) "a"'s
      *
      * then confirm that find() for the deepest name succeeds without any
      * disruption, and the resulting chain has the expected level.
-     * Note that "a." and the root name (".") belong to the same level.
-     * So the possible maximum level is MAX_LABELS - 1, not MAX_LABELS.
+     * Note that the root name (".") solely belongs to a single level,
+     * so the levels begin with 2.
      */
-    for (unsigned int i = 1; i < Name::MAX_LABELS; ++i) {
+    for (unsigned int i = 2; i <= Name::MAX_LABELS; ++i) {
         node_name = Name("a.").concatenate(node_name);
         EXPECT_EQ(RBTree<int>::SUCCESS, tree.insert(node_name, &rbtnode));
         RBTreeNodeChain<int> found_chain;
@@ -523,4 +523,44 @@ TEST_F(RBTreeTest, swap) {
     tree2.dumpTree(out);
     ASSERT_EQ(str1.str(), out.str());
 }
+
+// Matching in the "root zone" may be special (e.g. there's no parent,
+// any domain names should be considered a subdomain of it), so it makes
+// sense to test cases with the root zone explicitly.
+TEST_F(RBTreeTest, root) {
+    RBTree<int> root;
+    root.insert(Name::ROOT_NAME(), &rbtnode);
+    rbtnode->setData(RBNode<int>::NodeDataPtr(new int(1)));
+
+    EXPECT_EQ(RBTree<int>::EXACTMATCH,
+              root.find(Name::ROOT_NAME(), &crbtnode));
+    EXPECT_EQ(rbtnode, crbtnode);
+    EXPECT_EQ(RBTree<int>::PARTIALMATCH,
+              root.find(Name("example.com"), &crbtnode));
+    EXPECT_EQ(rbtnode, crbtnode);
+
+    // Insert a new name that better matches the query name.  find() should
+    // find the better one.
+    root.insert(Name("com"), &rbtnode);
+    rbtnode->setData(RBNode<int>::NodeDataPtr(new int(2)));
+    EXPECT_EQ(RBTree<int>::PARTIALMATCH,
+              root.find(Name("example.com"), &crbtnode));
+    EXPECT_EQ(rbtnode, crbtnode);
+
+    // Perform the same tests for the tree that allows matching against empty
+    // nodes.
+    RBTree<int> root_emptyok(true);
+    root_emptyok.insert(Name::ROOT_NAME(), &rbtnode);
+    EXPECT_EQ(RBTree<int>::EXACTMATCH,
+              root_emptyok.find(Name::ROOT_NAME(), &crbtnode));
+    EXPECT_EQ(rbtnode, crbtnode);
+    EXPECT_EQ(RBTree<int>::PARTIALMATCH,
+              root_emptyok.find(Name("example.com"), &crbtnode));
+    EXPECT_EQ(rbtnode, crbtnode);
+
+    root.insert(Name("com"), &rbtnode);
+    EXPECT_EQ(RBTree<int>::PARTIALMATCH,
+              root.find(Name("example.com"), &crbtnode));
+    EXPECT_EQ(rbtnode, crbtnode);
+}
 }

src/lib/datasrc/tests/test_datasrc.cc

@@ -273,6 +273,18 @@ const struct RRData nosoa_example_records[] = {
 };
 
 //
+// zone data for apexcname.example.
+//
+const struct RRData apexcname_example_records[] = {
+    {"apexcname.example", "CNAME", "canonical.apexcname.example"},
+    {"canonical.apexcname.example", "SOA",
+     "master.apexcname.example "
+     "admin.apexcname.example. 1234 3600 1800 2419200 7200"},
+    {NULL, NULL, NULL}
+};
+
+
+//
 // empty data set, for convenience.
 //
 const struct RRData empty_records[] = {
@@ -288,7 +300,8 @@ const struct ZoneData zone_data[] = {
     { "loop.example", "IN", loop_example_records, empty_records },
     { "nons.example", "IN", nons_example_records, empty_records },
     { "nons-dname.example", "IN", nonsdname_example_records, empty_records },
-    { "nosoa.example", "IN", nosoa_example_records, empty_records }
+    { "nosoa.example", "IN", nosoa_example_records, empty_records },
+    { "apexcname.example", "IN", nosoa_example_records, empty_records }
 };
 const size_t NUM_ZONES = sizeof(zone_data) / sizeof(zone_data[0]);
 

src/lib/log/Makefile.am

@@ -23,6 +23,11 @@ liblog_la_SOURCES += message_types.h
 liblog_la_SOURCES += root_logger_name.cc root_logger_name.h
 liblog_la_SOURCES += strutil.h strutil.cc
 
+EXTRA_DIST  = README
+EXTRA_DIST += messagedef.mes
+EXTRA_DIST += logger_impl_log4cxx.cc logger_impl_log4cxx.h
+EXTRA_DIST += xdebuglevel.cc xdebuglevel.h
+
 # Note: the ordering matters: -Wno-... must follow -Wextra (defined in
 # B10_CXXFLAGS)
 liblog_la_CXXFLAGS = $(AM_CXXFLAGS)

src/lib/log/documentation.txt

@@ -59,7 +59,7 @@ The steps in using the system are:
 
 2. Run it through the message compiler to produce the .h and .cc files.  It
    is intended that this step be included in the build process.  However,
-   for not run the compiler (found in the "compiler" subdirectory) manually.
+   for now run the compiler (found in the "compiler" subdirectory) manually.
    The only argument is the name of the message file: it will produce as
    output two files, having the same name as the input file but with file
    types of ".h" and ".cc".
@@ -74,7 +74,7 @@ The steps in using the system are:
 4. Declare loggers in your code and use them to log messages.  This is
    described in more detail below.
 
-5. To set the debug level and run-time message file, call runTimeInit (declared
+5. To set the debug level and run-time message file, call initLogger (declared
    in logger_support.h) in the main program unit.  This is a temporary solution
    for Year 2, and will be replaced at a later date, the information coming
    from the configuration database.
@@ -106,7 +106,7 @@ UNKNOWN     unknown message
 
 Points to note:
 * Leading and trailing space are trimmed from the line.  Although the above
-  exampl,e has every line starting at column 1, the lines could be indented
+  example has every line starting at column 1, the lines could be indented
   if desired.
 
 * Blank lines are ignored.
@@ -120,10 +120,8 @@ Points to note:
   * $PREFIX, which has one argument: the string used to prefix symbols.  If
     absent, there is no prefix to the symbols. (Prefixes are explained below.)
   * $NAMESPACE, which has one argument: the namespace in which the symbols are
-    created.  (Specifying the argument as a double colon - i.e. "$NAMESPACE
-    ::" puts the symbol definitions in the unnamed namespace.  And not
-    including a $NAMESPACE directive will result in the symbols note being
-    put in any namespace.
+    created.  In the absence of a $NAMESPACE directive, symbols will be put
+    in the global namespace.
 
 * Lines starting + indicate an explanation for the preceding message.  These
   are intended to be processed by a separate program and used to generate
@@ -151,13 +149,11 @@ The message compiler processes the message file to produce two files:
 the form:
 
    namespace <namespace> {
-   isc::log::MessageID PREFIX_IDENTIFIER = "IDENTIFIER";
+   extern const isc::log::MessageID PREFIX_IDENTIFIER;
       :
    }
 
-The symbols define the keys in the global message dictionary.  At present
-they are defined as std::strings, but a future implementation could redefine
-them as numeric values.
+The symbols define the keys in the global message dictionary.
 
 The namespace enclosing the symbols is set by the $NAMESPACE directive.
 
@@ -167,14 +163,18 @@ ABC with "MSG_" to give the symbol MSG_ABC.  Similarly "$PREFIX E" would
 prefix it with "E" to give the symbol EABC.  If no $PREFIX is given, no
 prefix appears (so the symbol in this example would be ABC).
 
-The header file also includes a couple of lines to ensure that the message
-text is included in the final program image.
+2) A C++ source file (called <message-file-name>.cc) that holds the definitions
+of the global symbols and code to insert the symbols and messages into the map.
 
+Symbols are defined to be equal to strings holding the identifier, e.g.
 
-2) A C++ source file (called <message-file-name>.cc) that holds the code to
-insert the symbols and messages into the map.
+   extern const isc::log::MessageID MSG_DUPLNS = "DUPLNS";
 
-This file declares an array of identifiers/messages in the form:
+(The implementation allows symbols to be compared.  However, use of strings
+should not be assumed - a future implementation may change this.)
+
+In addition, the file declares an array of identifiers/messages and an object
+to add them to the global dictionary:
 
     namespace {
     const char* values[] = {
@@ -183,20 +183,9 @@ This file declares an array of identifiers/messages in the form:
         :
         NULL
     };
-    }
-
-(A more complex structure to group identifiers and their messages could be
-imposed, but as the array is generated by code and will be read by code,
-it is not needed.)
 
-It then declares an object that will add information to the global dictionary:
-
-    MessageInitializer <message-file-name>_<time>(values);
-
-(Declaring the object as "static" or in the anonymous namespace runs the risk
-of it being optimised away when the module is compiled with optimisation.
-But giving it a standard name would cause a clash when multiple files are
-used, hence an attempt at disambiguation.)
+    const isc::log::MessageInitializer initializer(values);
+    }
 
 The constructor of the MessageInitializer object retrieves the singleton
 global Dictionary object (created using standard methods to avoid the
@@ -233,10 +222,10 @@ To use the current version of the logging:
 
        isc::log::Logger logger("myname", true);
 
-   The argument is ignored for underlying implementations other than log4cxx.
-   See below for the use of this argument.
+   (The argument is required to support a possible future implementation of
+   logging.  Currently it has no effect.)
 
-3. The main program unit should include a call to isc::log::runTimeInit()
+3. The main program unit should include a call to isc::log::initLogger()
    (defined in logger_support.h) to set the logging severity, debug log level,
    and external message file.
 
@@ -250,13 +239,13 @@ To use the current version of the logging:
         isc::log::NONE
 
    b) The debug log level is only interpreted when the severity is DEBUG and
-      is an integer raning from 0 to 99.  0 should be used for the highest-level
-      debug messages and 99 for the lowest-level (and typically more verbose)
-      messages.
+      is an integer ranging from 0 to 99.  0 should be used for the
+      highest-level debug messages and 99 for the lowest-level (and typically
+      more verbose) messages.
 
    c) Name of an external message file.  This is the same as a standard message
-      file, although it should not include the $PREFIX directive. (A single
-      $PREFIX directive will be ignored; multiple directives will cause the
+      file, although it should not include any directives. (A single directive
+      of a particular type will be ignored; multiple directives will cause the
       read of the file to fail with an error.)  If a message is replaced, the
       message should include the same printf-format directives in the same order
       as the original message.
@@ -340,7 +329,7 @@ Logging Sources v Logging Severities
 When logging events, make a distinction between events related to the server
 and events related to DNS messages received.  Caution needs to be exercised
 with the latter as, if the logging is enabled in the normal course of events,
-such logging could be a denoial of service vector. For example, suppose that
+such logging could be a denial of service vector. For example, suppose that
 the main authoritiative service logger were to log both zone loading and
 unloading as INFO and a warning message if it received an invalid packet. An
 attacker could make the INFO messages unusable by flooding the server with
@@ -353,7 +342,7 @@ DEBUG is not enabled by default, so these events will not be recorded unless
 DEBUG is specifically chosen.
 
 b) Record system-related and packet-related messages via different loggers
-(e.g.  in the example given, sever events could be logged using the logger
+(e.g.  in the example given, server events could be logged using the logger
 "auth" and packet-related events at that level logged using the logger
 "pkt-auth".)  As the loggers are independent and the severity levels
 independent, fine-tuning of what and what is not recorded can be achieved.
@@ -379,56 +368,9 @@ Outstanding Issues
 
 log4cxx Issues
 ==============
+Some experimental code to utilise log4cxx as an underlying implementation
+is present in the source code directory although it is not currently used.
+The files are:
 
-Second Argument in Logger Constructor
--------------------------------------
-As noted above, when using log4cxx as the underlying implementation, the
-argument to the logger's constructor should be set true if declaring the
-logger within a method and set false (or omitted) if declaring the logger
-external to an execution unit.
-
-This is due to an apparent bug in the underlying log4cxx, where the deletion
-of a statically-declared object at program termination can cause a segment
-fault. (The destruction of internal memory structures can sometimes happen
-out of order.)  By default the Logger class creates the structures in
-its constructor but does not delete them in the destruction.  The default
-behavious works because instead of reclaiming memory at program run-down,
-the operating system reclaims it when the process is deleted.
-
-Setting the second argument "true" causes the Logger's destructor to delete
-the log4cxx structures.  This does not cause a problem if the program is
-not terminating.  So use the second form when declaring an automatic instance
-of isc::log::Logger on the stack.
-
-Building with log4cxx
----------------------
-Owing to issues with versions of log4cxx on different systems, log4cxx was
-temporarily disabled.  To use log4cxx on your system:
-
-* Uncomment the log4cxx lines in configure.ac
-* In src/lib/log, replace the logger_impl.{cc,h} files with their log4cxx
-  equivalents, i.e.
-
-  cp logger_impl_log4cxx.h logger_impl.h
-  cp logger_impl_log4cxx.cc logger_impl.cc
-
-* In src/lib/log/Makefile.am, uncomment the lines:
-
-  # AM_CPPFLAGS += $(LOG4CXX_INCLUDES)
-
-  # liblog_la_SOURCES += xdebuglevel.cc xdebuglevel.h
-
-  # liblog_la_LDFLAGS = $(LOG4CXX_LDFLAGS)
-
-* In src/lib/log/test, re-enable testing of the log4cxx implementation
-  class, i.e.
-
-  cp logger_impl_log4cxx_unittest.cc logger_impl_unittest.cc
-
-  ... and uncomment the following lines in Makefile.am:
-
-  # run_unittests_SOURCES += logger_impl_unittest.cc
-
-  # run_unittests_SOURCES += xdebuglevel_unittest.cc
-
-Then rebuild the system from scratch.
+   logger_impl_log4cxx.{cc,h}
+   xdebuglevel.{cc,h}

src/lib/nsas/address_entry.h

@@ -21,7 +21,7 @@
 /// convenience methods for accessing and updating the information.
 
 #include <stdint.h>
-#include "asiolink.h"
+#include <asiolink/io_address.h>
 
 namespace isc {
 namespace nsas {
@@ -39,7 +39,7 @@ public:
     {}
 
     /// \return Address object
-    asiolink::IOAddress getAddress() const {
+    const asiolink::IOAddress& getAddress() const {
         return address_;
     }
 

src/lib/nsas/asiolink.h

@@ -18,41 +18,4 @@
 #include <string>
 #include <sys/socket.h>
 
-namespace asiolink {
-
-/// \brief IO Address Dummy Class
-///
-/// As part of ther resolver, Evan has written the asiolink.h file, which
-/// encapsulates some of the boost::asio classes.  Until these are checked
-/// into trunk and merged with this branch, these dummy classes should fulfill
-/// their function.
-
-class IOAddress {
-public:
-    /// \param address_str String representing the address
-    IOAddress(const std::string& address_str) : address_(address_str)
-    {}
-
-    /// \param Just a virtual destructor
-    virtual ~ IOAddress() { }
-
-    /// \return Textual representation of the address
-    std::string toText() const
-    {return address_;}
-
-    /// \return Address family of the address
-    virtual short getFamily() const {
-        return ((address_.find(".") != std::string::npos) ? AF_INET : AF_INET6);
-    }
-
-    /// \return true if two addresses are equal
-    bool equal(const IOAddress& address) const
-    {return (toText() == address.toText());}
-
-private:
-    std::string     address_;       ///< Address represented
-};
-
-}   // namespace asiolink
-
 #endif // __ASIOLINK_H

src/lib/nsas/nameserver_address.cc

@@ -23,7 +23,9 @@ namespace nsas {
 void
 NameserverAddress::updateRTT(uint32_t rtt) const {
     // We delegate it to the address entry inside the nameserver entry
-    ns_->updateAddressRTT(rtt, address_.getAddress(), family_);
+    if (ns_) {
+        ns_->updateAddressRTT(rtt, address_.getAddress(), family_);
+    }
 }
 
 } // namespace nsas

src/lib/nsas/nameserver_address_store.cc

@@ -93,5 +93,18 @@ NameserverAddressStore::lookup(const string& zone, const RRClass& class_code,
     zone_obj.second->addCallback(callback, family);
 }
 
+void
+NameserverAddressStore::cancel(const string& zone,
+    const RRClass& class_code,
+    const boost::shared_ptr<AddressRequestCallback>& callback,
+    AddressFamily family)
+{
+    boost::shared_ptr<ZoneEntry> entry(zone_hash_->get(HashKey(zone,
+                                                               class_code)));
+    if (entry) {
+        entry->removeCallback(callback, family);
+    }
+}
+
 } // namespace nsas
 } // namespace isc

src/lib/nsas/nameserver_address_store.h

@@ -87,6 +87,13 @@ public:
         boost::shared_ptr<AddressRequestCallback> callback, AddressFamily
         family = ANY_OK);
 
+    /// \brief cancel the given lookup action
+    ///
+    /// \param callback Callback object that would be called
+    void cancel(const std::string& zone, const dns::RRClass& class_code,
+                const boost::shared_ptr<AddressRequestCallback>& callback,
+                AddressFamily family = ANY_OK);
+
     /// \brief Protected Members
     ///
     /// These members should be private.  However, with so few public methods

src/lib/nsas/nameserver_entry.cc

@@ -35,6 +35,8 @@
 #include <dns/question.h>
 #include <resolve/resolver_interface.h>
 
+#include <asiolink/io_address.h>
+
 #include "address_entry.h"
 #include "nameserver_address.h"
 #include "nameserver_entry.h"
@@ -140,7 +142,7 @@ NameserverEntry::setAddressRTT(const IOAddress& address, uint32_t rtt) {
     AddressFamily family(V4_ONLY);
     for (;;) {
         BOOST_FOREACH(AddressEntry& entry, addresses_[family]) {
-            if (entry.getAddress().equal(address)) {
+            if (entry.getAddress().equals(address)) {
                 entry.setRTT(rtt);
                 return;
             }
@@ -181,7 +183,7 @@ NameserverEntry::updateAddressRTT(uint32_t rtt,
 {
     Lock lock(mutex_);
     for (size_t i(0); i < addresses_[family].size(); ++ i) {
-        if (addresses_[family][i].getAddress().equal(address)) {
+        if (addresses_[family][i].getAddress().equals(address)) {
             updateAddressRTTAtIndex(rtt, i, family);
             return;
         }
@@ -226,8 +228,9 @@ class NameserverEntry::ResolverCallback :
                 response_message->getRcode() != isc::dns::Rcode::NOERROR() ||
                 response_message->getRRCount(isc::dns::Message::SECTION_ANSWER) == 0) {
                 failureInternal(lock);
+                return;
             }
-                
+            
             isc::dns::RRsetIterator rrsi =
                 response_message->beginSection(isc::dns::Message::SECTION_ANSWER);
             const isc::dns::RRsetPtr response = *rrsi;

src/lib/nsas/tests/Makefile.am

@@ -54,6 +54,7 @@ endif
 
 run_unittests_LDADD += $(top_builddir)/src/lib/nsas/libnsas.la
 run_unittests_LDADD += $(top_builddir)/src/lib/dns/libdns++.la
+run_unittests_LDADD += $(top_builddir)/src/lib/asiolink/libasiolink.la
 run_unittests_LDADD += $(top_builddir)/src/lib/exceptions/libexceptions.la
 endif
 

src/lib/nsas/tests/address_entry_unittest.cc

@@ -24,12 +24,12 @@
 #include <stdint.h>
 
 
-#include "../asiolink.h"
+#include <asiolink/io_address.h>
 #include "../address_entry.h"
 
 static std::string V4A_TEXT("1.2.3.4");
 static std::string V4B_TEXT("5.6.7.8");
-static std::string V6A_TEXT("2001:dead:beef::0");
+static std::string V6A_TEXT("2001:dead:beef::");
 static std::string V6B_TEXT("1984:1985::1986:1987");
 
 using namespace asiolink;

src/lib/nsas/tests/nameserver_address_unittest.cc

@@ -100,7 +100,7 @@ protected:
 
 // Test that the address is equal to the address in NameserverEntry
 TEST_F(NameserverAddressTest, Address) {
-    EXPECT_TRUE(ns_address_.getAddress().equal( ns_sample_.getAddressAtIndex(TEST_ADDRESS_INDEX)));
+    EXPECT_TRUE(ns_address_.getAddress().equals( ns_sample_.getAddressAtIndex(TEST_ADDRESS_INDEX)));
 
     boost::shared_ptr<NameserverEntry> empty_ne((NameserverEntry*)NULL);
     // It will throw an NullNameserverEntryPointer exception with the empty NameserverEntry shared pointer

src/lib/nsas/tests/nameserver_entry_unittest.cc

@@ -153,7 +153,7 @@ TEST_F(NameserverEntryTest, SetRTT) {
     int matchcount = 0;
     for (NameserverEntry::AddressVectorIterator i = newvec.begin();
         i != newvec.end(); ++i) {
-        if (i->getAddress().equal(first_address)) {
+        if (i->getAddress().equals(first_address)) {
             ++matchcount;
             EXPECT_EQ(i->getAddressEntry().getRTT(), new_rtt);
         }
@@ -188,7 +188,7 @@ TEST_F(NameserverEntryTest, Unreachable) {
     int matchcount = 0;
     for (NameserverEntry::AddressVectorIterator i = newvec.begin();
         i != newvec.end(); ++i) {
-        if (i->getAddress().equal(first_address)) {
+        if (i->getAddress().equals(first_address)) {
             ++matchcount;
             EXPECT_TRUE(i->getAddressEntry().isUnreachable());
         }

src/lib/nsas/tests/zone_entry_unittest.cc

@@ -165,9 +165,9 @@ protected:
         EXPECT_EQ(failure_count, callback_->unreachable_count_);
         EXPECT_EQ(success_count, callback_->successes_.size());
         for (size_t i = 0; i < callback_->successes_.size(); ++ i) {
-            EXPECT_TRUE(IOAddress("192.0.2.1").equal(
+            EXPECT_TRUE(IOAddress("192.0.2.1").equals(
                 callback_->successes_[i].getAddress()) ||
-                IOAddress("2001:db8::1").equal(
+                IOAddress("2001:db8::1").equals(
                 callback_->successes_[i].getAddress()));
         }
     }
@@ -234,7 +234,7 @@ TEST_F(ZoneEntryTest, ChangedNS) {
     EXPECT_NO_THROW(resolver_->answer(1, ns_name_, RRType::A(),
         rdata::in::A("192.0.2.1")));
     ASSERT_EQ(1, callback_->successes_.size());
-    EXPECT_TRUE(IOAddress("192.0.2.1").equal(
+    EXPECT_TRUE(IOAddress("192.0.2.1").equals(
         callback_->successes_[0].getAddress()));
     EXPECT_NO_THROW(resolver_->answer(2, ns_name_, RRType::AAAA(),
         rdata::in::AAAA("2001:db8::1")));
@@ -257,7 +257,7 @@ TEST_F(ZoneEntryTest, ChangedNS) {
     EXPECT_NO_THROW(resolver_->answer(4, different_name, RRType::A(),
         rdata::in::A("192.0.2.2")));
     ASSERT_EQ(2, callback_->successes_.size());
-    EXPECT_TRUE(IOAddress("192.0.2.2").equal(
+    EXPECT_TRUE(IOAddress("192.0.2.2").equals(
         callback_->successes_[1].getAddress()));
 
     // And now, switch back, as it timed out again
@@ -270,7 +270,7 @@ TEST_F(ZoneEntryTest, ChangedNS) {
     EXPECT_EQ(7, resolver_->requests.size());
     EXPECT_EQ(Fetchable::READY, zone->getState());
     ASSERT_EQ(3, callback_->successes_.size());
-    EXPECT_TRUE(IOAddress("192.0.2.1").equal(
+    EXPECT_TRUE(IOAddress("192.0.2.1").equals(
         callback_->successes_[0].getAddress()));
 }
 
@@ -306,9 +306,9 @@ TEST_F(ZoneEntryTest, CallbacksAnswered) {
          rdata::in::A("192.0.2.1")));
     // Two are answered (ANY and V4)
     ASSERT_EQ(2, callback_->successes_.size());
-    EXPECT_TRUE(IOAddress("192.0.2.1").equal(
+    EXPECT_TRUE(IOAddress("192.0.2.1").equals(
         callback_->successes_[0].getAddress()));
-    EXPECT_TRUE(IOAddress("192.0.2.1").equal(
+    EXPECT_TRUE(IOAddress("192.0.2.1").equals(
         callback_->successes_[1].getAddress()));
     // None are rejected
     EXPECT_EQ(0, callback_->unreachable_count_);
@@ -318,7 +318,7 @@ TEST_F(ZoneEntryTest, CallbacksAnswered) {
     // This should answer the third callback
     EXPECT_EQ(0, callback_->unreachable_count_);
     ASSERT_EQ(3, callback_->successes_.size());
-    EXPECT_TRUE(IOAddress("2001:db8::1").equal(
+    EXPECT_TRUE(IOAddress("2001:db8::1").equals(
         callback_->successes_[2].getAddress()));
     // It should think it is ready
     EXPECT_EQ(Fetchable::READY, zone->getState());
@@ -326,7 +326,7 @@ TEST_F(ZoneEntryTest, CallbacksAnswered) {
     zone->addCallback(callback_, V4_ONLY);
     EXPECT_EQ(3, resolver_->requests.size());
     ASSERT_EQ(4, callback_->successes_.size());
-    EXPECT_TRUE(IOAddress("192.0.2.1").equal(
+    EXPECT_TRUE(IOAddress("192.0.2.1").equals(
         callback_->successes_[3].getAddress()));
     EXPECT_EQ(0, callback_->unreachable_count_);
 }
@@ -366,9 +366,9 @@ TEST_F(ZoneEntryTest, CallbacksAOnly) {
     EXPECT_NO_THROW(resolver_->answer(1, ns_name_, RRType::A(),
         rdata::in::A("192.0.2.1")));
     ASSERT_EQ(2, callback_->successes_.size());
-    EXPECT_TRUE(IOAddress("192.0.2.1").equal(
+    EXPECT_TRUE(IOAddress("192.0.2.1").equals(
         callback_->successes_[0].getAddress()));
-    EXPECT_TRUE(IOAddress("192.0.2.1").equal(
+    EXPECT_TRUE(IOAddress("192.0.2.1").equals(
         callback_->successes_[1].getAddress()));
     EXPECT_EQ(1, callback_->unreachable_count_);
     // Everything arriwed, so we are ready
@@ -377,7 +377,7 @@ TEST_F(ZoneEntryTest, CallbacksAOnly) {
     zone->addCallback(callback_, V4_ONLY);
     EXPECT_EQ(3, resolver_->requests.size());
     ASSERT_EQ(3, callback_->successes_.size());
-    EXPECT_TRUE(IOAddress("192.0.2.1").equal(
+    EXPECT_TRUE(IOAddress("192.0.2.1").equals(
         callback_->successes_[2].getAddress()));
     EXPECT_EQ(1, callback_->unreachable_count_);
 
@@ -439,9 +439,9 @@ TEST_F(ZoneEntryTest, CallbackTwoNS) {
     // The other callbacks should be answered now
     EXPECT_EQ(2, callback_->unreachable_count_);
     ASSERT_EQ(2, callback_->successes_.size());
-    EXPECT_TRUE(IOAddress("2001:db8::1").equal(
+    EXPECT_TRUE(IOAddress("2001:db8::1").equals(
         callback_->successes_[0].getAddress()));
-    EXPECT_TRUE(IOAddress("2001:db8::1").equal(
+    EXPECT_TRUE(IOAddress("2001:db8::1").equals(
         callback_->successes_[1].getAddress()));
 }
 
@@ -534,7 +534,7 @@ TEST_F(ZoneEntryTest, AddressTimeout) {
          rdata::in::A("192.0.2.1"), 0));
     // It answers, not rejects
     ASSERT_EQ(1, callback_->successes_.size());
-    EXPECT_TRUE(IOAddress("192.0.2.1").equal(
+    EXPECT_TRUE(IOAddress("192.0.2.1").equals(
         callback_->successes_[0].getAddress()));
     EXPECT_EQ(0, callback_->unreachable_count_);
     // As well with IPv6
@@ -551,7 +551,7 @@ TEST_F(ZoneEntryTest, AddressTimeout) {
          rdata::in::A("192.0.2.1"), 0));
     EXPECT_EQ(0, callback_->unreachable_count_);
     ASSERT_EQ(2, callback_->successes_.size());
-    EXPECT_TRUE(IOAddress("192.0.2.1").equal(
+    EXPECT_TRUE(IOAddress("192.0.2.1").equals(
         callback_->successes_[1].getAddress()));
 }
 

src/lib/nsas/zone_entry.cc

@@ -143,7 +143,7 @@ class ZoneEntry::ResolverCallback :
                         Name ns_name(dynamic_cast<const rdata::generic::NS&>(
                             iterator->getCurrent()).getNSName());
                         // Try to find it in the old ones
-                        map<string, NameserverPtr>::iterator old_ns(old.find(
+                        std::map<string, NameserverPtr>::iterator old_ns(old.find(
                             ns_name.toText()));
                         /*
                          * We didn't have this nameserver before. So we just
@@ -261,6 +261,23 @@ ZoneEntry::addCallback(CallbackPtr callback, AddressFamily family) {
     }
 }
 
+void
+ZoneEntry::removeCallback(const CallbackPtr& callback, AddressFamily family) {
+    Lock lock(mutex_);
+    std::vector<boost::shared_ptr<AddressRequestCallback> >::iterator i = 
+        callbacks_[family].begin();
+    for (; i != callbacks_[family].end(); ++i) {
+        if (*i == callback) {
+            callbacks_[family].erase(i);
+            // At this point, a callback should only be in the list
+            // once (enforced by RunningQuery doing only one at a time)
+            // If that changes, we need to revise this (can't delete
+            // elements from a list we're looping over)
+            return;
+        }
+    }
+}
+
 namespace {
 
 // This just moves items from one container to another

src/lib/nsas/zone_entry.h

@@ -101,6 +101,15 @@ public:
     void addCallback(boost::shared_ptr<AddressRequestCallback>
         callback, AddressFamily family);
 
+    /**
+     * \short Remove a callback from the list
+     *
+     * \param callback The callback itself.
+     * \param family Which address family is acceptable as an answer?
+     */
+    void removeCallback(const boost::shared_ptr<AddressRequestCallback>&
+                        callback, AddressFamily family);
+
     /// \short Protected members, so they can be accessed by tests.
     //@{
 protected:

src/lib/resolve/Makefile.am

@@ -14,5 +14,16 @@ libresolve_la_SOURCES = resolve.h resolve.cc
 libresolve_la_SOURCES += resolver_interface.h
 libresolve_la_SOURCES += resolver_callback.h resolver_callback.cc
 libresolve_la_SOURCES += response_classifier.cc response_classifier.h
+libresolve_la_SOURCES += recursive_query.cc recursive_query.h
 libresolve_la_LIBADD = $(top_builddir)/src/lib/dns/libdns++.la
 libresolve_la_LIBADD += $(top_builddir)/src/lib/exceptions/libexceptions.la
+
+# Note: the ordering matters: -Wno-... must follow -Wextra (defined in
+# B10_CXXFLAGS)
+libresolve_la_CXXFLAGS = $(AM_CXXFLAGS)
+if USE_CLANGPP
+# For clang++, we need to turn off -Werror completely.
+libresolve_la_CXXFLAGS += -Wno-error
+endif
+libresolve_la_CPPFLAGS = $(AM_CPPFLAGS)
+

src/lib/resolve/recursive_query.cc

@@ -0,0 +1,757 @@
+// Copyright (C) 2011  Internet Systems Consortium, Inc. ("ISC")
+//
+// Permission to use, copy, modify, and/or distribute this software for any
+// purpose with or without fee is hereby granted, provided that the above
+// copyright notice and this permission notice appear in all copies.
+//
+// THE SOFTWARE IS PROVIDED "AS IS" AND ISC DISCLAIMS ALL WARRANTIES WITH
+// REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
+// AND FITNESS.  IN NO EVENT SHALL ISC BE LIABLE FOR ANY SPECIAL, DIRECT,
+// INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
+// LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
+// OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
+// PERFORMANCE OF THIS SOFTWARE.
+
+#include <netinet/in.h>
+#include <stdlib.h>
+#include <sys/socket.h>
+#include <unistd.h>             // for some IPC/network system calls
+
+#include <boost/lexical_cast.hpp>
+#include <boost/bind.hpp>
+
+#include <config.h>
+
+#include <log/dummylog.h>
+
+#include <dns/question.h>
+#include <dns/message.h>
+#include <dns/opcode.h>
+
+#include <resolve/resolve.h>
+#include <cache/resolver_cache.h>
+#include <nsas/address_request_callback.h>
+#include <nsas/nameserver_address.h>
+
+#include <asio.hpp>
+#include <asiolink/dns_service.h>
+#include <asiolink/io_fetch.h>
+#include <asiolink/io_service.h>
+#include <resolve/recursive_query.h>
+
+using isc::log::dlog;
+using namespace isc::dns;
+
+namespace asiolink {
+
+typedef std::vector<std::pair<std::string, uint16_t> > AddressVector;
+
+// Here we do not use the typedef above, as the SunStudio compiler
+// mishandles this in its name mangling, and wouldn't compile.
+// We can probably use a typedef, but need to move it to a central
+// location and use it consistently.
+RecursiveQuery::RecursiveQuery(DNSService& dns_service,
+    isc::nsas::NameserverAddressStore& nsas,
+    isc::cache::ResolverCache& cache,
+    const std::vector<std::pair<std::string, uint16_t> >& upstream,
+    const std::vector<std::pair<std::string, uint16_t> >& upstream_root,
+    int query_timeout, int client_timeout, int lookup_timeout,
+    unsigned retries) :
+    dns_service_(dns_service),
+    nsas_(nsas), cache_(cache),
+    upstream_(new AddressVector(upstream)),
+    upstream_root_(new AddressVector(upstream_root)),
+    test_server_("", 0),
+    query_timeout_(query_timeout), client_timeout_(client_timeout),
+    lookup_timeout_(lookup_timeout), retries_(retries)
+{
+}
+
+// Set the test server - only used for unit testing.
+
+void
+RecursiveQuery::setTestServer(const std::string& address, uint16_t port) {
+    dlog("Setting test server to " + address + "(" +
+            boost::lexical_cast<std::string>(port) + ")");
+    test_server_.first = address;
+    test_server_.second = port;
+}
+
+
+namespace {
+
+typedef std::pair<std::string, uint16_t> addr_t;
+
+/*
+ * This is a query in progress. When a new query is made, this one holds
+ * the context information about it, like how many times we are allowed
+ * to retry on failure, what to do when we succeed, etc.
+ *
+ * Used by RecursiveQuery::sendQuery.
+ */
+class RunningQuery : public IOFetch::Callback {
+
+class ResolverNSASCallback : public isc::nsas::AddressRequestCallback {
+public:
+    ResolverNSASCallback(RunningQuery* rq) : rq_(rq) {}
+    
+    void success(const isc::nsas::NameserverAddress& address) {
+        dlog("Found a nameserver, sending query to " + address.getAddress().toText());
+        rq_->nsasCallbackCalled();
+        rq_->sendTo(address);
+    }
+    
+    void unreachable() {
+        dlog("Nameservers unreachable");
+        // Drop query or send servfail?
+        rq_->nsasCallbackCalled();
+        rq_->makeSERVFAIL();
+        rq_->callCallback(true);
+        rq_->stop();
+    }
+
+private:
+    RunningQuery* rq_;
+};
+
+
+private:
+    // The io service to handle async calls
+    IOService& io_;
+
+    // Info for (re)sending the query (the question and destination)
+    Question question_;
+
+    // This is where we build and store our final answer
+    MessagePtr answer_message_;
+
+    // currently we use upstream as the current list of NS records
+    // we should differentiate between forwarding and resolving
+    boost::shared_ptr<AddressVector> upstream_;
+
+    // Test server - only used for testing.  This takes precedence over all
+    // other servers if the port is non-zero.
+    std::pair<std::string, uint16_t> test_server_;
+
+    // Buffer to store the intermediate results.
+    OutputBufferPtr buffer_;
+
+    // The callback will be called when we have either decided we
+    // are done, or when we give up
+    isc::resolve::ResolverInterface::CallbackPtr resolvercallback_;
+
+    // Protocol used for the last query.  This is set to IOFetch::UDP when a
+    // new upstream query is initiated, and changed to IOFetch::TCP if a
+    // packet is returned with the TC bit set.  It is stored here to detect the
+    // case of a TCP packet being returned with the TC bit set.
+    IOFetch::Protocol protocol_;
+
+    // To prevent both unreasonably long cname chains and cname loops,
+    // we simply keep a counter of the number of CNAMEs we have
+    // followed so far (and error if it exceeds RESOLVER_MAX_CNAME_CHAIN
+    // from lib/resolve/response_classifier.h)
+    unsigned cname_count_;
+
+    /*
+     * TODO Do something more clever with timeouts. In the long term, some
+     *     computation of average RTT, increase with each retry, etc.
+     */
+    // Timeout information for outgoing queries
+    int query_timeout_;
+    unsigned retries_;
+
+    // normal query state
+
+    // Update the question that will be sent to the server
+    void setQuestion(const Question& new_question) {
+        question_ = new_question;
+    }
+
+    // TODO: replace by our wrapper
+    asio::deadline_timer client_timer;
+    asio::deadline_timer lookup_timer;
+
+    // If we timed out ourselves (lookup timeout), stop issuing queries
+    bool done_;
+
+    // If we have a client timeout, we call back with a failure message,
+    // but we do not stop yet. We use this variable to make sure we 
+    // don't call back a second time later
+    bool callback_called_;
+
+    // Reference to our NSAS
+    isc::nsas::NameserverAddressStore& nsas_;
+
+    // Reference to our cache
+    isc::cache::ResolverCache& cache_;
+    
+    // the 'current' zone we are in (i.e.) we start out at the root,
+    // and for each delegation this gets updated with the zone the
+    // delegation points to.
+    // TODO: make this a Name (it is a string right now because most
+    // of the call we use it in take a string, we need update those
+    // too).
+    std::string cur_zone_;
+    
+    // This is the handler we pass on to the NSAS; it is called when
+    // the NSAS has an address for us to query
+    boost::shared_ptr<ResolverNSASCallback> nsas_callback_;
+
+    // this is set to true if we have asked the nsas to give us
+    // an address and we are waiting for it to call us back.
+    // We use is to cancel the outstanding callback in case we
+    // have a lookup timeout and decide to give up
+    bool nsas_callback_out_;
+
+    // This is the nameserver we have an outstanding query to.
+    // It is used to update the RTT once the query returns
+    isc::nsas::NameserverAddress current_ns_address;
+
+    // The moment in time we sent a query to the nameserver above.
+    struct timeval current_ns_qsent_time;
+    
+    // RunningQuery deletes itself when it is done. In order for us
+    // to do this safely, we must make sure that there are no events
+    // that might call back to it. There are two types of events in
+    // this sense; the timers we set ourselves (lookup and client),
+    // and outstanding queries to nameservers. When each of these is
+    // started, we increase this value. When they fire, it is decreased
+    // again. We cannot delete ourselves until this value is back to 0.
+    //
+    // Note that the NSAS callback is *not* seen as an outstanding
+    // event; we can cancel the NSAS callback safely.
+    size_t outstanding_events_;
+
+    // perform a single lookup; first we check the cache to see
+    // if we have a response for our query stored already. if
+    // so, call handlerecursiveresponse(), if not, we call send()
+    void doLookup() {
+        dlog("doLookup: try cache");
+        Message cached_message(Message::RENDER);
+        isc::resolve::initResponseMessage(question_, cached_message);
+        if (cache_.lookup(question_.getName(), question_.getType(),
+                          question_.getClass(), cached_message)) {
+            dlog("Message found in cache, continuing with that");
+            // Should these be set by the cache too?
+            cached_message.setOpcode(Opcode::QUERY());
+            cached_message.setRcode(Rcode::NOERROR());
+            cached_message.setHeaderFlag(Message::HEADERFLAG_QR);
+            if (handleRecursiveAnswer(cached_message)) {
+                callCallback(true);
+                stop();
+            }
+        } else {
+            cur_zone_ = ".";
+            send();
+        }
+        
+    }
+
+    // Send the current question to the given nameserver address
+    void sendTo(const isc::nsas::NameserverAddress& address) {
+        // We need to keep track of the Address, so that we can update
+        // the RTT
+        current_ns_address = address;
+        gettimeofday(&current_ns_qsent_time, NULL);
+        ++outstanding_events_;
+        IOFetch query(protocol_, io_, question_,
+            current_ns_address.getAddress(),
+            53, buffer_, this,
+            query_timeout_);
+        io_.get_io_service().post(query);
+    }
+    
+    // 'general' send; if we are in forwarder mode, send a query to
+    // a random nameserver in our forwarders list. If we are in
+    // recursive mode, ask the NSAS to give us an address.
+    void send(IOFetch::Protocol protocol = IOFetch::UDP) {
+        // If are in forwarder mode, send it to a random
+        // forwarder. If not, ask the NSAS for an address
+        const int uc = upstream_->size();
+        protocol_ = protocol;   // Store protocol being used for this
+        if (test_server_.second != 0) {
+            dlog("Sending upstream query (" + question_.toText() +
+                 ") to test server at " + test_server_.first);
+            ++outstanding_events_;
+            IOFetch query(protocol, io_, question_,
+                test_server_.first,
+                test_server_.second, buffer_, this,
+                query_timeout_);
+            io_.get_io_service().post(query);
+        } else if (uc > 0) {
+            // TODO: use boost, or rand()-utility function we provide
+            int serverIndex = rand() % uc;
+            dlog("Sending upstream query (" + question_.toText() +
+                ") to " + upstream_->at(serverIndex).first);
+            ++outstanding_events_;
+            IOFetch query(protocol, io_, question_,
+                upstream_->at(serverIndex).first,
+                upstream_->at(serverIndex).second, buffer_, this,
+                query_timeout_);
+            io_.get_io_service().post(query);
+        } else {
+            // Ask the NSAS for an address for the current zone,
+            // the callback will call the actual sendTo()
+            dlog("Look up nameserver for " + cur_zone_ + " in NSAS");
+            // Can we have multiple calls to nsas_out? Let's assume not
+            // for now
+            assert(!nsas_callback_out_);
+            nsas_callback_out_ = true;
+            nsas_.lookup(cur_zone_, question_.getClass(), nsas_callback_);
+        }
+    }
+    
+    // Called by our NSAS callback handler so we know we do not have
+    // an outstanding NSAS call anymore.
+    void nsasCallbackCalled() {
+        nsas_callback_out_ = false;
+    }
+
+    // This function is called by operator() and lookup();
+    // We have an answer either from a nameserver or the cache, and
+    // we do not know yet if this is a final answer we can send back or
+    // that more recursive processing needs to be done.
+    // Depending on the content, we go on recursing or return
+    //
+    // This method also updates the cache, depending on the content
+    // of the message
+    //
+    // returns true if we are done (either we have an answer or an
+    //              error message)
+    // returns false if we are not done
+    bool handleRecursiveAnswer(const Message& incoming) {
+        dlog("Handle response");
+        // In case we get a CNAME, we store the target
+        // here (classify() will set it when it walks through
+        // the cname chain to verify it).
+        Name cname_target(question_.getName());
+        
+        isc::resolve::ResponseClassifier::Category category =
+            isc::resolve::ResponseClassifier::classify(
+                question_, incoming, cname_target, cname_count_);
+
+        bool found_ns_address = false;
+            
+        switch (category) {
+        case isc::resolve::ResponseClassifier::ANSWER:
+        case isc::resolve::ResponseClassifier::ANSWERCNAME:
+            // Done. copy and return.
+            dlog("Response is an answer");
+            isc::resolve::copyResponseMessage(incoming, answer_message_);
+            cache_.update(*answer_message_);
+            return true;
+            break;
+        case isc::resolve::ResponseClassifier::CNAME:
+            dlog("Response is CNAME!");
+            // (unfinished) CNAME. We set our question_ to the CNAME
+            // target, then start over at the beginning (for now, that
+            // is, we reset our 'current servers' to the root servers).
+            if (cname_count_ >= RESOLVER_MAX_CNAME_CHAIN) {
+                // just give up
+                dlog("CNAME chain too long");
+                makeSERVFAIL();
+                return true;
+            }
+
+            answer_message_->appendSection(Message::SECTION_ANSWER,
+                                           incoming);
+
+            question_ = Question(cname_target, question_.getClass(),
+                                 question_.getType());
+
+            dlog("Following CNAME chain to " + question_.toText());
+            doLookup();
+            return false;
+            break;
+        case isc::resolve::ResponseClassifier::NXDOMAIN:
+        case isc::resolve::ResponseClassifier::NXRRSET:
+            dlog("Response is NXDOMAIN or NXRRSET");
+            // NXDOMAIN, just copy and return.
+            dlog(incoming.toText());
+            isc::resolve::copyResponseMessage(incoming, answer_message_);
+            // no negcache yet
+            //cache_.update(*answer_message_);
+            return true;
+            break;
+        case isc::resolve::ResponseClassifier::REFERRAL:
+            dlog("Response is referral");
+            cache_.update(incoming);
+            // Referral. For now we just take the first glue address
+            // we find and continue with that
+
+            // auth section should have at least one RRset
+            // and one of them should be an NS (otherwise
+            // classifier should have error'd)
+            // TODO: should we check if it really is subzone?
+            for (RRsetIterator rrsi = incoming.beginSection(Message::SECTION_AUTHORITY);
+                 rrsi != incoming.endSection(Message::SECTION_AUTHORITY) && !found_ns_address;
+                 ++rrsi) {
+                ConstRRsetPtr rrs = *rrsi;
+                if (rrs->getType() == RRType::NS()) {
+                    // TODO: make cur_zone_ a Name instead of a string
+                    // (this requires a few API changes in related
+                    // libraries, so as not to need many conversions)
+                    cur_zone_ = rrs->getName().toText();
+                    dlog("Referred to zone " + cur_zone_);
+                    found_ns_address = true;
+                    break;
+                }
+            }
+
+            if (found_ns_address) {
+                // next resolver round
+                // we do NOT use doLookup() here, but send() (i.e. we
+                // skip the cache), since if we had the final answer
+                // instead of a delegation cached, we would have been
+                // there by now.
+                send();
+                return false;
+            } else {
+                dlog("No NS RRset in referral?");
+                // TODO this will result in answering with the delegation. oh well
+                isc::resolve::copyResponseMessage(incoming, answer_message_);
+                return true;
+            }
+            break;
+        case isc::resolve::ResponseClassifier::TRUNCATED:
+            // Truncated packet.  If the protocol we used for the last one is
+            // UDP, re-query using TCP.  Otherwise regard it as an error.
+            if (protocol_ == IOFetch::UDP) {
+                dlog("Response truncated, re-querying over TCP");
+                send(IOFetch::TCP);
+                return false;
+            }
+            // Was a TCP query so we have received a packet over TCP with the TC
+            // bit set: drop through to common error processing.
+            // TODO: Can we use what we have received instead of discarding it?
+
+        case isc::resolve::ResponseClassifier::EMPTY:
+        case isc::resolve::ResponseClassifier::EXTRADATA:
+        case isc::resolve::ResponseClassifier::INVNAMCLASS:
+        case isc::resolve::ResponseClassifier::INVTYPE:
+        case isc::resolve::ResponseClassifier::MISMATQUEST:
+        case isc::resolve::ResponseClassifier::MULTICLASS:
+        case isc::resolve::ResponseClassifier::NOTONEQUEST:
+        case isc::resolve::ResponseClassifier::NOTRESPONSE:
+        case isc::resolve::ResponseClassifier::NOTSINGLE:
+        case isc::resolve::ResponseClassifier::OPCODE:
+        case isc::resolve::ResponseClassifier::RCODE:
+            // Should we try a different server rather than SERVFAIL?
+            makeSERVFAIL();
+            return true;
+            break;
+        }
+
+        // Since we do not have a default in the switch above,
+        // the compiler should have errored on any missing case
+        // statements.
+        assert(false);
+        return true;
+    }
+    
+public:
+    RunningQuery(IOService& io,
+        const Question& question,
+        MessagePtr answer_message,
+        boost::shared_ptr<AddressVector> upstream,
+        std::pair<std::string, uint16_t>& test_server,
+        OutputBufferPtr buffer,
+        isc::resolve::ResolverInterface::CallbackPtr cb,
+        int query_timeout, int client_timeout, int lookup_timeout,
+        unsigned retries,
+        isc::nsas::NameserverAddressStore& nsas,
+        isc::cache::ResolverCache& cache) :
+        io_(io),
+        question_(question),
+        answer_message_(answer_message),
+        upstream_(upstream),
+        test_server_(test_server),
+        buffer_(buffer),
+        resolvercallback_(cb),
+        protocol_(IOFetch::UDP),
+        cname_count_(0),
+        query_timeout_(query_timeout),
+        retries_(retries),
+        client_timer(io.get_io_service()),
+        lookup_timer(io.get_io_service()),
+        done_(false),
+        callback_called_(false),
+        nsas_(nsas),
+        cache_(cache),
+        nsas_callback_(new ResolverNSASCallback(this)),
+        nsas_callback_out_(false),
+        outstanding_events_(0)
+    {
+        // Setup the timer to stop trying (lookup_timeout)
+        if (lookup_timeout >= 0) {
+            lookup_timer.expires_from_now(
+                boost::posix_time::milliseconds(lookup_timeout));
+            ++outstanding_events_;
+            lookup_timer.async_wait(boost::bind(&RunningQuery::lookupTimeout, this));
+        }
+        
+        // Setup the timer to send an answer (client_timeout)
+        if (client_timeout >= 0) {
+            client_timer.expires_from_now(
+                boost::posix_time::milliseconds(client_timeout));
+            ++outstanding_events_;
+            client_timer.async_wait(boost::bind(&RunningQuery::clientTimeout, this));
+        }
+        
+        doLookup();
+    }
+
+    // called if we have a lookup timeout; if our callback has
+    // not been called, call it now. Then stop.
+    void lookupTimeout() {
+        if (!callback_called_) {
+            makeSERVFAIL();
+            callCallback(true);
+        }
+        assert(outstanding_events_ > 0);
+        --outstanding_events_;
+        stop();
+    }
+    
+    // called if we have a client timeout; if our callback has
+    // not been called, call it now. But do not stop.
+    void clientTimeout() {
+        if (!callback_called_) {
+            makeSERVFAIL();
+            callCallback(true);
+        }
+        assert(outstanding_events_ > 0);
+        --outstanding_events_;
+        if (outstanding_events_ == 0) {
+            stop();
+        }
+    }
+
+    // If the callback has not been called yet, call it now
+    // If success is true, we call 'success' with our answer_message
+    // If it is false, we call failure()
+    void callCallback(bool success) {
+        if (!callback_called_) {
+            callback_called_ = true;
+
+            // There are two types of messages we could store in the
+            // cache;
+            // 1. answers to our fetches from authoritative servers,
+            //    exactly as we receive them, and
+            // 2. answers to queries we received from clients, which
+            //    have received additional processing (following CNAME
+            //    chains, for instance)
+            //
+            // Doing only the first would mean we would have to re-do
+            // processing when we get data from our cache, and doing
+            // only the second would miss out on the side-effect of
+            // having nameserver data in our cache.
+            //
+            // So right now we do both. Since the cache (currently)
+            // stores Messages on their question section only, this
+            // does mean that we overwrite the messages we stored in
+            // the previous iteration if we are following a delegation.
+            if (success) {
+                resolvercallback_->success(answer_message_);
+            } else {
+                resolvercallback_->failure();
+            }
+        }
+    }
+
+    // We are done. If there are no more outstanding events, we delete
+    // ourselves. If there are any, we do not.
+    void stop() {
+        done_ = true;
+        if (nsas_callback_out_) {
+            nsas_.cancel(cur_zone_, question_.getClass(), nsas_callback_);
+            nsas_callback_out_ = false;
+        }
+        client_timer.cancel();
+        lookup_timer.cancel();
+        if (outstanding_events_ > 0) {
+            return;
+        } else {
+            delete this;
+        }
+    }
+
+    // This function is used as callback from DNSQuery.
+    virtual void operator()(IOFetch::Result result) {
+        // XXX is this the place for TCP retry?
+        assert(outstanding_events_ > 0);
+        --outstanding_events_;
+        
+        if (!done_ && result != IOFetch::TIME_OUT) {
+            // we got an answer
+
+            // Update the NSAS with the time it took
+            struct timeval cur_time;
+            gettimeofday(&cur_time, NULL);
+            uint32_t rtt;
+            if (cur_time.tv_sec >= current_ns_qsent_time.tv_sec ||
+                cur_time.tv_usec > current_ns_qsent_time.tv_usec) {
+                rtt = 1000 * (cur_time.tv_sec - current_ns_qsent_time.tv_sec);
+                rtt += (cur_time.tv_usec - current_ns_qsent_time.tv_usec) / 1000;
+            } else {
+                rtt = 1;
+            }
+
+            dlog("RTT: " + boost::lexical_cast<std::string>(rtt));
+            current_ns_address.updateRTT(rtt);
+            
+            Message incoming(Message::PARSE);
+            InputBuffer ibuf(buffer_->getData(), buffer_->getLength());
+            incoming.fromWire(ibuf);
+
+            buffer_->clear();
+            if (recursive_mode() &&
+                incoming.getRcode() == Rcode::NOERROR()) {
+                done_ = handleRecursiveAnswer(incoming);
+            } else {
+                isc::resolve::copyResponseMessage(incoming, answer_message_);
+                done_ = true;
+            }
+            
+            if (done_) {
+                callCallback(true);
+                stop();
+            }
+        } else if (!done_ && retries_--) {
+            // Query timed out, but we have some retries, so send again
+            dlog("Timeout for " + question_.toText() + " to " + current_ns_address.getAddress().toText() + ", resending query");
+            if (recursive_mode()) {
+                current_ns_address.updateRTT(isc::nsas::AddressEntry::UNREACHABLE);
+            }
+            send();
+        } else {
+            // We are either already done, or out of retries
+            if (recursive_mode() && result == IOFetch::TIME_OUT) {
+                dlog("Timeout for " + question_.toText() + " to " + current_ns_address.getAddress().toText() + ", giving up");
+                current_ns_address.updateRTT(isc::nsas::AddressEntry::UNREACHABLE);
+            }
+            if (!callback_called_) {
+                makeSERVFAIL();
+                callCallback(true);
+            }
+            stop();
+        }
+    }
+    
+    // Clear the answer parts of answer_message, and set the rcode
+    // to servfail
+    void makeSERVFAIL() {
+        isc::resolve::makeErrorMessage(answer_message_, Rcode::SERVFAIL());
+    }
+    
+    // Returns true if we are in 'recursive' mode
+    // Returns false if we are in 'forwarding' mode
+    // (i.e. if we have anything in upstream_)
+    bool recursive_mode() const {
+        return upstream_->empty();
+    }
+};
+
+}
+
+void
+RecursiveQuery::resolve(const QuestionPtr& question,
+    const isc::resolve::ResolverInterface::CallbackPtr callback)
+{
+    IOService& io = dns_service_.getIOService();
+
+    MessagePtr answer_message(new Message(Message::RENDER));
+    isc::resolve::initResponseMessage(*question, *answer_message);
+
+    OutputBufferPtr buffer(new OutputBuffer(0));
+
+    dlog("Asked to resolve: " + question->toText());
+    
+    dlog("Try out cache first (direct call to resolve)");
+    // First try to see if we have something cached in the messagecache
+    if (cache_.lookup(question->getName(), question->getType(),
+                      question->getClass(), *answer_message) &&
+        answer_message->getRRCount(Message::SECTION_ANSWER) > 0) {
+        dlog("Message found in cache, returning that");
+        // TODO: err, should cache set rcode as well?
+        answer_message->setRcode(Rcode::NOERROR());
+        callback->success(answer_message);
+    } else {
+        // Perhaps we only have the one RRset?
+        // TODO: can we do this? should we check for specific types only?
+        RRsetPtr cached_rrset = cache_.lookup(question->getName(),
+                                              question->getType(),
+                                              question->getClass());
+        if (cached_rrset) {
+            dlog("Found single RRset in cache");
+            answer_message->addRRset(Message::SECTION_ANSWER,
+                                     cached_rrset);
+            answer_message->setRcode(Rcode::NOERROR());
+            callback->success(answer_message);
+        } else {
+            dlog("Message not found in cache, starting recursive query");
+            // It will delete itself when it is done
+            new RunningQuery(io, *question, answer_message, upstream_,
+                             test_server_, buffer, callback,
+                             query_timeout_, client_timeout_,
+                             lookup_timeout_, retries_, nsas_, cache_);
+        }
+    }
+}
+
+void
+RecursiveQuery::resolve(const Question& question,
+                        MessagePtr answer_message,
+                        OutputBufferPtr buffer,
+                        DNSServer* server)
+{
+    // XXX: eventually we will need to be able to determine whether
+    // the message should be sent via TCP or UDP, or sent initially via
+    // UDP and then fall back to TCP on failure, but for the moment
+    // we're only going to handle UDP.
+    IOService& io = dns_service_.getIOService();
+
+    isc::resolve::ResolverInterface::CallbackPtr crs(
+        new isc::resolve::ResolverCallbackServer(server));
+
+    // TODO: general 'prepareinitialanswer'
+    answer_message->setOpcode(isc::dns::Opcode::QUERY());
+    answer_message->addQuestion(question);
+    
+    dlog("Asked to resolve: " + question.toText());
+    
+    // First try to see if we have something cached in the messagecache
+    dlog("Try out cache first (started by incoming event)");
+    if (cache_.lookup(question.getName(), question.getType(),
+                      question.getClass(), *answer_message) &&
+        answer_message->getRRCount(Message::SECTION_ANSWER) > 0) {
+        dlog("Message found in cache, returning that");
+        // TODO: err, should cache set rcode as well?
+        answer_message->setRcode(Rcode::NOERROR());
+        crs->success(answer_message);
+    } else {
+        // Perhaps we only have the one RRset?
+        // TODO: can we do this? should we check for specific types only?
+        RRsetPtr cached_rrset = cache_.lookup(question.getName(),
+                                              question.getType(),
+                                              question.getClass());
+        if (cached_rrset) {
+            dlog("Found single RRset in cache");
+            answer_message->addRRset(Message::SECTION_ANSWER,
+                                     cached_rrset);
+            answer_message->setRcode(Rcode::NOERROR());
+            crs->success(answer_message);
+        } else {
+            dlog("Message not found in cache, starting recursive query");
+            // It will delete itself when it is done
+            new RunningQuery(io, question, answer_message, upstream_,
+                             test_server_, buffer, crs, query_timeout_,
+                             client_timeout_, lookup_timeout_, retries_,
+                             nsas_, cache_);
+        }
+    }
+}
+
+
+
+} // namespace asiolink

src/lib/asiolink/recursive_query.h

@@ -12,12 +12,13 @@
 // OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
 // PERFORMANCE OF THIS SOFTWARE.
 
-#ifndef __ASIOLINK_RECURSIVE_QUERY_H
-#define __ASIOLINK_RECURSIVE_QUERY_H 1
+#ifndef __RECURSIVE_QUERY_H
+#define __RECURSIVE_QUERY_H 1
 
 #include <asiolink/dns_service.h>
 #include <asiolink/dns_server.h>
 #include <dns/buffer.h>
+#include <nsas/nameserver_address_store.h>
 #include <cache/resolver_cache.h>
 
 namespace asiolink {
@@ -52,6 +53,8 @@ public:
     /// \param retries how many times we try again (0 means just send and
     ///     and return if it returs).
     RecursiveQuery(DNSService& dns_service,
+                   isc::nsas::NameserverAddressStore& nsas,
+                   isc::cache::ResolverCache& cache,
                    const std::vector<std::pair<std::string, uint16_t> >&
                    upstream, 
                    const std::vector<std::pair<std::string, uint16_t> >&
@@ -98,20 +101,33 @@ public:
                  isc::dns::MessagePtr answer_message,
                  isc::dns::OutputBufferPtr buffer,
                  DNSServer* server);
+
+    /// \brief Set Test Server
+    ///
+    /// This method is *only* for unit testing the class.  If set, it enables
+    /// recursive behaviour but, regardless of responses received, sends every
+    /// query to the test server.
+    ///
+    /// The test server is enabled by setting a non-zero port number.
+    ///
+    /// \param address IP address of the test server.
+    /// \param port Port number of the test server
+    void setTestServer(const std::string& address, uint16_t port);
+    
 private:
     DNSService& dns_service_;
+    isc::nsas::NameserverAddressStore& nsas_;
+    isc::cache::ResolverCache& cache_;
     boost::shared_ptr<std::vector<std::pair<std::string, uint16_t> > >
         upstream_;
     boost::shared_ptr<std::vector<std::pair<std::string, uint16_t> > >
         upstream_root_;
+    std::pair<std::string, uint16_t> test_server_;
     int query_timeout_;
     int client_timeout_;
     int lookup_timeout_;
     unsigned retries_;
-    // Cache. TODO: I think we want this initialized in Resolver class,
-    // not here
-    isc::cache::ResolverCache cache_;
 };
 
 }      // namespace asiolink
-#endif // __ASIOLINK_RECURSIVE_QUERY_H
+#endif // __RECURSIVE_QUERY_H

src/lib/resolve/response_classifier.cc

@@ -114,13 +114,17 @@ ResponseClassifier::Category ResponseClassifier::classify(
             );
 
     // If there is nothing in the answer section, it is a referral - unless
-    // there is nothing in the authority section
+    // there is no NS in the authority section
     if (answer.empty()) {
         if (authority.empty()) {
             return (EMPTY);
-        } else {
-            return (REFERRAL);
         }
+        for (int i = 0; i < authority.size(); ++i) {
+            if (authority[i]->getType() == RRType::NS()) {
+                return (REFERRAL);
+            }
+        }
+        return (NXRRSET);
     }
 
     // Look at two cases - one RRset in the answer and multiple RRsets in

src/lib/resolve/response_classifier.h

@@ -53,6 +53,7 @@ public:
         ANSWERCNAME,        ///< Response was a CNAME chain ending in an answer
         CNAME,              ///< Response was a CNAME
         NXDOMAIN,           ///< Response was an NXDOMAIN
+        NXRRSET,            ///< Response was name exists, but type does not
         REFERRAL,           ///< Response contains a referral
 
         // Codes indicating that a message is invalid.  Note that the error()

src/lib/resolve/tests/Makefile.am

@@ -14,12 +14,19 @@ TESTS += run_unittests
 run_unittests_CPPFLAGS = $(AM_CPPFLAGS) $(GTEST_INCLUDES)
 run_unittests_LDFLAGS = $(AM_LDFLAGS) $(GTEST_LDFLAGS)
 run_unittests_SOURCES = run_unittests.cc
+run_unittests_SOURCES += $(top_srcdir)/src/lib/dns/tests/unittest_util.h
+run_unittests_SOURCES += $(top_srcdir)/src/lib/dns/tests/unittest_util.cc
 run_unittests_SOURCES += resolve_unittest.cc
 run_unittests_SOURCES += resolver_callback_unittest.cc
 run_unittests_SOURCES += response_classifier_unittest.cc
+run_unittests_SOURCES += recursive_query_unittest.cc
+run_unittests_SOURCES += recursive_query_unittest_2.cc
 
 run_unittests_LDADD = $(GTEST_LDADD)
 run_unittests_LDADD +=  $(top_builddir)/src/lib/exceptions/libexceptions.la
+run_unittests_LDADD +=  $(top_builddir)/src/lib/nsas/libnsas.la
+run_unittests_LDADD +=  $(top_builddir)/src/lib/cache/libcache.la
+run_unittests_LDADD +=  $(top_builddir)/src/lib/asiolink/libasiolink.la
 run_unittests_LDADD +=  $(top_builddir)/src/lib/resolve/libresolve.la
 run_unittests_LDADD +=  $(top_builddir)/src/lib/dns/libdns++.la
 

src/lib/asiolink/tests/recursive_query_unittest.cc

@@ -33,6 +33,9 @@
 #include <dns/buffer.h>
 #include <dns/message.h>
 
+#include <nsas/nameserver_address_store.h>
+#include <cache/resolver_cache.h>
+
 // IMPORTANT: We shouldn't directly use ASIO definitions in this test.
 // In particular, we must not include asio.hpp in this file.
 // The asiolink module is primarily intended to be a wrapper that hide the
@@ -41,7 +44,7 @@
 // if we include asio.hpp unless we specify a special compiler option.
 // If we need to test something at the level of underlying ASIO and need
 // their definition, that test should go to asiolink/internal/tests.
-#include <asiolink/recursive_query.h>
+#include <resolve/recursive_query.h>
 #include <asiolink/io_socket.h>
 #include <asiolink/io_service.h>
 #include <asiolink/io_message.h>
@@ -343,6 +346,12 @@ protected:
         private:
             bool* done_;
     };
+    
+    class MockResolver : public isc::resolve::ResolverInterface {
+        void resolve(const QuestionPtr& question,
+                     const ResolverInterface::CallbackPtr& callback) {
+        }
+    };
 
     // This version of mock server just stops the io_service when it is resumed
     // the second time. (Used in the clientTimeout test, where resume
@@ -403,6 +412,8 @@ protected:
     // need to recreate a new one within one onstance of this class
     IOService* io_service_;
     DNSService* dns_service_;
+    isc::nsas::NameserverAddressStore* nsas_;
+    isc::cache::ResolverCache cache_;
     ASIOCallBack* callback_;
     int callback_protocol_;
     int callback_native_;
@@ -418,6 +429,8 @@ RecursiveQueryTest::RecursiveQueryTest() :
 {
     io_service_ = new IOService();
     setDNSService(true, true);
+    boost::shared_ptr<MockResolver>mock_resolver(new MockResolver());
+    nsas_ = new isc::nsas::NameserverAddressStore(mock_resolver);
 }
 
 TEST_F(RecursiveQueryTest, v6UDPSend) {
@@ -519,6 +532,7 @@ TEST_F(RecursiveQueryTest, recursiveSetupV4) {
     setDNSService(true, false);
     uint16_t port = boost::lexical_cast<uint16_t>(TEST_CLIENT_PORT);
     EXPECT_NO_THROW(RecursiveQuery(*dns_service_,
+                                   *nsas_, cache_,
                                    singleAddress(TEST_IPV4_ADDR, port),
                                    singleAddress(TEST_IPV4_ADDR, port)));
 }
@@ -527,6 +541,7 @@ TEST_F(RecursiveQueryTest, recursiveSetupV6) {
     setDNSService(false, true);
     uint16_t port = boost::lexical_cast<uint16_t>(TEST_CLIENT_PORT);
     EXPECT_NO_THROW(RecursiveQuery(*dns_service_,
+                                   *nsas_, cache_,
                                    singleAddress(TEST_IPV6_ADDR, port),
                                    singleAddress(TEST_IPV6_ADDR,port)));
 }
@@ -545,6 +560,7 @@ TEST_F(RecursiveQueryTest, forwarderSend) {
 
     MockServer server(*io_service_);
     RecursiveQuery rq(*dns_service_,
+                      *nsas_, cache_,
                       singleAddress(TEST_IPV4_ADDR, port),
                       singleAddress(TEST_IPV4_ADDR, port));
 
@@ -632,6 +648,7 @@ TEST_F(RecursiveQueryTest, forwardQueryTimeout) {
     // Do the answer
     const uint16_t port = boost::lexical_cast<uint16_t>(TEST_CLIENT_PORT);
     RecursiveQuery query(*dns_service_,
+                         *nsas_, cache_,
                          singleAddress(TEST_IPV4_ADDR, port),
                          singleAddress(TEST_IPV4_ADDR, port),
                          10, 4000, 3000, 2);
@@ -649,8 +666,8 @@ TEST_F(RecursiveQueryTest, forwardQueryTimeout) {
     int num = 0;
     bool read_success = tryRead(sock_, recv_options, 3, &num);
 
-    // The query should fail
-    EXPECT_FALSE(done);
+    // The query should 'succeed' with an error response
+    EXPECT_TRUE(done);
     EXPECT_EQ(3, num);
     EXPECT_TRUE(read_success);
 }
@@ -666,8 +683,7 @@ TEST_F(RecursiveQueryTest, forwardClientTimeout) {
 
     // Prepare the server
     bool done1(true);
-    bool done2(true);
-    MockServerStop2 server(*io_service_, &done1, &done2);
+    MockServerStop server(*io_service_, &done1);
 
     MessagePtr answer(new Message(Message::RENDER));
 
@@ -677,6 +693,7 @@ TEST_F(RecursiveQueryTest, forwardClientTimeout) {
     // Since the lookup timer has not fired, it should retry
     // four times
     RecursiveQuery query(*dns_service_,
+                         *nsas_, cache_,
                          singleAddress(TEST_IPV4_ADDR, port),
                          singleAddress(TEST_IPV4_ADDR, port),
                          200, 480, 4000, 4);
@@ -690,15 +707,14 @@ TEST_F(RecursiveQueryTest, forwardClientTimeout) {
     // we know it'll fail, so make it a shorter timeout
     int recv_options = setSocketTimeout(sock_, 1, 0);
 
-    // Try to read 5 times
+    // Try to read 4 times
     int num = 0;
-    bool read_success = tryRead(sock_, recv_options, 5, &num);
+    bool read_success = tryRead(sock_, recv_options, 4, &num);
 
-    // The query should fail, but we should have kept on trying
+    // The query should fail
     EXPECT_TRUE(done1);
-    EXPECT_FALSE(done2);
-    EXPECT_EQ(5, num);
-    EXPECT_TRUE(read_success);
+    EXPECT_EQ(3, num);
+    EXPECT_FALSE(read_success);
 }
 
 // If we set lookup timeout to lower than querytimeout*retries, we should
@@ -721,6 +737,7 @@ TEST_F(RecursiveQueryTest, forwardLookupTimeout) {
     // Set up the test so that it will retry 5 times, but the lookup
     // timeout will fire after only 3 normal timeouts
     RecursiveQuery query(*dns_service_,
+                         *nsas_, cache_,
                          singleAddress(TEST_IPV4_ADDR, port),
                          singleAddress(TEST_IPV4_ADDR, port),
                          200, 4000, 480, 5);
@@ -737,12 +754,55 @@ TEST_F(RecursiveQueryTest, forwardLookupTimeout) {
     int num = 0;
     bool read_success = tryRead(sock_, recv_options, 5, &num);
 
-    // The query should fail
-    EXPECT_FALSE(done);
+    // The query should fail and respond with an error
+    EXPECT_TRUE(done);
     EXPECT_EQ(3, num);
     EXPECT_FALSE(read_success);
 }
 
+// Set everything very low and see if this doesn't cause weird
+// behaviour
+TEST_F(RecursiveQueryTest, lowtimeouts) {
+    // Prepare the service (we do not use the common setup, we do not answer
+    setDNSService();
+
+    // Prepare the socket
+    sock_ = createTestSocket();
+
+    // Prepare the server
+    bool done(true);
+    MockServerStop server(*io_service_, &done);
+
+    MessagePtr answer(new Message(Message::RENDER));
+
+    // Do the answer
+    const uint16_t port = boost::lexical_cast<uint16_t>(TEST_CLIENT_PORT);
+    // Set up the test so that it will retry 5 times, but the lookup
+    // timeout will fire after only 3 normal timeouts
+    RecursiveQuery query(*dns_service_,
+                         *nsas_, cache_,
+                         singleAddress(TEST_IPV4_ADDR, port),
+                         singleAddress(TEST_IPV4_ADDR, port),
+                         1, 1, 1, 1);
+    Question question(Name("example.net"), RRClass::IN(), RRType::A());
+    OutputBufferPtr buffer(new OutputBuffer(0));
+    query.resolve(question, answer, buffer, &server);
+
+    // Run the test
+    io_service_->run();
+
+    int recv_options = setSocketTimeout(sock_, 1, 0);
+
+    // Try to read 5 times, should stop after 3 reads
+    int num = 0;
+    bool read_success = tryRead(sock_, recv_options, 5, &num);
+
+    // The query should fail and respond with an error
+    EXPECT_TRUE(done);
+    EXPECT_EQ(1, num);
+    EXPECT_FALSE(read_success);
+}
+
 // as mentioned above, we need a more better framework for this,
 // in addition to that, this sends out queries into the world
 // (which we should catch somehow and fake replies for)
@@ -755,7 +815,8 @@ TEST_F(RecursiveQueryTest, DISABLED_recursiveSendOk) {
     
     MockServerStop server(*io_service_, &done);
     vector<pair<string, uint16_t> > empty_vector;
-    RecursiveQuery rq(*dns_service_, empty_vector, empty_vector, 10000, 0);
+    RecursiveQuery rq(*dns_service_, *nsas_, cache_, empty_vector,
+                      empty_vector, 10000, 0);
 
     Question q(Name("www.isc.org"), RRClass::IN(), RRType::A());
     OutputBufferPtr buffer(new OutputBuffer(0));
@@ -780,7 +841,8 @@ TEST_F(RecursiveQueryTest, DISABLED_recursiveSendNXDOMAIN) {
     
     MockServerStop server(*io_service_, &done);
     vector<pair<string, uint16_t> > empty_vector;
-    RecursiveQuery rq(*dns_service_, empty_vector, empty_vector, 10000, 0);
+    RecursiveQuery rq(*dns_service_, *nsas_, cache_, empty_vector,
+                      empty_vector, 10000, 0);
 
     Question q(Name("wwwdoesnotexist.isc.org"), RRClass::IN(), RRType::A());
     OutputBufferPtr buffer(new OutputBuffer(0));
@@ -793,4 +855,7 @@ TEST_F(RecursiveQueryTest, DISABLED_recursiveSendNXDOMAIN) {
     EXPECT_EQ(0, answer->getRRCount(Message::SECTION_ANSWER));
 }
 
+// TODO: add tests that check whether the cache is updated on succesfull
+// responses, and not updated on failures.
+
 }

src/lib/resolve/tests/recursive_query_unittest_2.cc

@@ -0,0 +1,656 @@
+// Copyright (C) 2011  Internet Systems Consortium, Inc. ("ISC")
+//
+// Permission to use, copy, modify, and/or distribute this software for any
+// purpose with or without fee is hereby granted, provided that the above
+// copyright notice and this permission notice appear in all copies.
+//
+// THE SOFTWARE IS PROVIDED "AS IS" AND ISC DISCLAIMS ALL WARRANTIES WITH
+// REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
+// AND FITNESS.  IN NO EVENT SHALL ISC BE LIABLE FOR ANY SPECIAL, DIRECT,
+// INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
+// LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
+// OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
+// PERFORMANCE OF THIS SOFTWARE.
+
+#include <algorithm>
+#include <cstdlib>
+#include <iomanip>
+#include <iostream>
+#include <string>
+
+#include <gtest/gtest.h>
+#include <boost/bind.hpp>
+
+
+#include <asio.hpp>
+
+#include <dns/buffer.h>
+#include <dns/question.h>
+#include <dns/message.h>
+#include <dns/messagerenderer.h>
+#include <dns/opcode.h>
+#include <dns/name.h>
+#include <dns/rcode.h>
+#include <dns/rrtype.h>
+#include <dns/rrset.h>
+#include <dns/rrttl.h>
+#include <dns/rdata.h>
+
+#include <asiolink/asiolink_utilities.h>
+#include <asiolink/dns_service.h>
+#include <asiolink/io_address.h>
+#include <asiolink/io_endpoint.h>
+#include <asiolink/io_fetch.h>
+#include <asiolink/io_service.h>
+#include <resolve/recursive_query.h>
+#include <resolve/resolver_interface.h>
+
+using namespace asio;
+using namespace asio::ip;
+using namespace isc::dns;
+using namespace isc::dns::rdata;
+using namespace isc::resolve;
+using namespace std;
+
+/// RecursiveQuery Test - 2
+///
+/// The second part of the RecursiveQuery unit tests, this attempts to get the
+/// RecursiveQuery object to follow a set of referrals for "www.example.org" to
+/// and to invoke TCP fallback on one of the queries.  In particular, we expect
+/// that the test will do the following in an attempt to resolve
+/// www.example.org:
+///
+/// - Send question over UDP to "root" - get referral to "org".
+/// - Send question over UDP to "org" - get referral to "example.org" with TC bit set.
+/// - Send question over TCP to "org" - get referral to "example.org".
+/// - Send question over UDP to "example.org" - get response for www.example.org.
+///
+/// (The order of queries is set in this way in order to also test that after a
+/// failover to TCP, queries revert to UDP).
+///
+/// By using the "test_server_" element of RecursiveQuery, all queries are
+/// directed to one or other of the "servers" in the RecursiveQueryTest2 class,
+/// regardless of the glue returned in referrals.
+
+namespace asiolink {
+
+const std::string TEST_ADDRESS = "127.0.0.1";   ///< Servers are on this address
+const uint16_t TEST_PORT = 5301;                ///< ... and this port
+const size_t BUFFER_SIZE = 1024;                ///< For all buffers
+const char* WWW_EXAMPLE_ORG = "192.0.2.254";    ///< Address of www.example.org
+
+// As the test is fairly long and complex, debugging "print" statements have
+// been left in although they are disabled.  Set the following to "true" to
+// enable them.
+const bool DEBUG_PRINT = false;
+
+class MockResolver : public isc::resolve::ResolverInterface {
+    void resolve(const QuestionPtr& question,
+                 const ResolverInterface::CallbackPtr& callback) {
+    }
+};
+
+
+
+/// \brief Test fixture for the RecursiveQuery Test
+class RecursiveQueryTest2 : public virtual ::testing::Test
+{
+public:
+
+    /// \brief Status of query
+    ///
+    /// Set before the query and then by each "server" when responding.
+    enum QueryStatus {
+        NONE = 0,                   ///< Default
+        UDP_ROOT = 1,               ///< Query root server over UDP
+        UDP_ORG = 2,                ///< Query ORG server over UDP
+        TCP_ORG = 3,                ///< Query ORG server over TCP
+        UDP_EXAMPLE_ORG = 4,        ///< Query EXAMPLE.ORG server over UDP
+        COMPLETE = 5                ///< Query is complete
+    };
+
+    // Common stuff
+    bool            debug_;                     ///< Set true for debug print
+    IOService       service_;                   ///< Service to run everything
+    DNSService      dns_service_;               ///< Resolver is part of "server"
+    QuestionPtr     question_;                  ///< What to ask
+    QueryStatus     last_;                      ///< What was the last state
+    QueryStatus     expected_;                  ///< Expected next state
+    OutputBufferPtr question_buffer_;           ///< Question we expect to receive
+    isc::nsas::NameserverAddressStore* nsas_;
+    isc::cache::ResolverCache cache_;
+
+    // Data for TCP Server
+    size_t          tcp_cumulative_;            ///< Cumulative TCP data received
+    tcp::endpoint   tcp_endpoint_;              ///< Endpoint for TCP receives
+    size_t          tcp_length_;                ///< Expected length value
+    uint8_t         tcp_receive_buffer_[BUFFER_SIZE];   ///< Receive buffer for TCP I/O
+    OutputBufferPtr tcp_send_buffer_;           ///< Send buffer for TCP I/O
+    tcp::socket     tcp_socket_;                ///< Socket used by TCP server
+
+    /// Data for UDP
+    udp::endpoint   udp_remote_;                ///< Endpoint for UDP receives
+    size_t          udp_length_;                ///< Expected length value
+    uint8_t         udp_receive_buffer_[BUFFER_SIZE];   ///< Receive buffer for UDP I/O
+    OutputBufferPtr udp_send_buffer_;           ///< Send buffer for UDP I/O
+    udp::socket     udp_socket_;                ///< Socket used by UDP server
+
+    /// \brief Constructor
+    RecursiveQueryTest2() :
+        debug_(DEBUG_PRINT),
+        service_(),
+        dns_service_(service_, NULL, NULL, NULL),
+        question_(new Question(Name("www.example.org"), RRClass::IN(), RRType::A())),
+        last_(NONE),
+        expected_(NONE),
+        question_buffer_(new OutputBuffer(BUFFER_SIZE)),
+        tcp_cumulative_(0),
+        tcp_endpoint_(asio::ip::address::from_string(TEST_ADDRESS), TEST_PORT),
+        tcp_length_(0),
+        tcp_receive_buffer_(),
+        tcp_send_buffer_(new OutputBuffer(BUFFER_SIZE)),
+        tcp_socket_(service_.get_io_service()),
+        udp_remote_(),
+        udp_length_(0),
+        udp_receive_buffer_(),
+        udp_send_buffer_(new OutputBuffer(BUFFER_SIZE)),
+        udp_socket_(service_.get_io_service(), udp::v4())
+    {
+        boost::shared_ptr<MockResolver>mock_resolver(new MockResolver());
+        nsas_ = new isc::nsas::NameserverAddressStore(mock_resolver);
+    }
+
+    /// \brief Set Common Message Bits
+    ///
+    /// Sets up the common bits of a response message returned by the handlers.
+    ///
+    /// \param msg Message buffer in RENDER mode.
+    /// \param qid QIT to set the message to
+    void setCommonMessage(isc::dns::Message& msg, uint16_t qid = 0) {
+        msg.setQid(qid);
+        msg.setHeaderFlag(Message::HEADERFLAG_QR);
+        msg.setOpcode(Opcode::QUERY());
+        msg.setHeaderFlag(Message::HEADERFLAG_AA);
+        msg.setRcode(Rcode::NOERROR());
+        msg.addQuestion(*question_);
+    }
+
+    /// \brief Set Referral to "org"
+    ///
+    /// Sets up the passed-in message (expected to be in "RENDER" mode to
+    /// indicate a referral to fictitious .org nameservers.
+    ///
+    /// \param msg Message to update with referral information.
+    void setReferralOrg(isc::dns::Message& msg) {
+        if (debug_) {
+            cout << "setReferralOrg(): creating referral to .org nameservers" << endl;
+        }
+
+        // Do a referral to org.  We'll define all NS records as "in-zone"
+        // nameservers (and supply glue) to avoid the possibility of the
+        // resolver starting another recursive query to resolve the address of
+        // a nameserver.
+        RRsetPtr org_ns(new RRset(Name("org."), RRClass::IN(), RRType::NS(), RRTTL(300)));
+        org_ns->addRdata(createRdata(RRType::NS(), RRClass::IN(), "ns1.org."));
+        org_ns->addRdata(createRdata(RRType::NS(), RRClass::IN(), "ns2.org."));
+        msg.addRRset(Message::SECTION_AUTHORITY, org_ns);
+
+        RRsetPtr org_ns1(new RRset(Name("ns1.org."), RRClass::IN(), RRType::A(), RRTTL(300)));
+        org_ns1->addRdata(createRdata(RRType::A(), RRClass::IN(), "192.0.2.1"));
+        msg.addRRset(Message::SECTION_ADDITIONAL, org_ns1);
+
+        RRsetPtr org_ns2(new RRset(Name("ns2.org."), RRClass::IN(), RRType::A(), RRTTL(300)));
+        org_ns2->addRdata(createRdata(RRType::A(), RRClass::IN(), "192.0.2.2"));
+        msg.addRRset(Message::SECTION_ADDITIONAL, org_ns2);
+    }
+
+    /// \brief Set Referral to "example.org"
+    ///
+    /// Sets up the passed-in message (expected to be in "RENDER" mode to
+    /// indicate a referral to fictitious example.org nameservers.
+    ///
+    /// \param msg Message to update with referral information.
+    void setReferralExampleOrg(isc::dns::Message& msg) {
+        if (debug_) {
+            cout << "setReferralExampleOrg(): creating referral to example.org nameservers" << endl;
+        }
+
+        // Do a referral to example.org.  As before, we'll define all NS
+        // records as "in-zone" nameservers (and supply glue) to avoid the
+        // possibility of the resolver starting another recursive query to look
+        // up the address of the nameserver.
+        RRsetPtr example_org_ns(new RRset(Name("example.org."), RRClass::IN(), RRType::NS(), RRTTL(300)));
+        example_org_ns->addRdata(createRdata(RRType::NS(), RRClass::IN(), "ns1.example.org."));
+        example_org_ns->addRdata(createRdata(RRType::NS(), RRClass::IN(), "ns2.example.org."));
+        msg.addRRset(Message::SECTION_AUTHORITY, example_org_ns);
+
+        RRsetPtr example_org_ns1(new RRset(Name("ns1.example.org."), RRClass::IN(), RRType::A(), RRTTL(300)));
+        example_org_ns1->addRdata(createRdata(RRType::A(), RRClass::IN(), "192.0.2.11"));
+        msg.addRRset(Message::SECTION_ADDITIONAL, example_org_ns1);
+
+        RRsetPtr example_org_ns2(new RRset(Name("ns2.example.org."), RRClass::IN(), RRType::A(), RRTTL(300)));
+        example_org_ns2->addRdata(createRdata(RRType::A(), RRClass::IN(), "192.0.2.21"));
+        msg.addRRset(Message::SECTION_ADDITIONAL, example_org_ns2);
+    }
+
+    /// \brief Set Answer to "www.example.org"
+    ///
+    /// Sets up the passed-in message (expected to be in "RENDER" mode) to
+    /// indicate an authoritative answer to www.example.org.
+    ///
+    /// \param msg Message to update with referral information.
+    void setAnswerWwwExampleOrg(isc::dns::Message& msg) {
+        if (debug_) {
+            cout << "setAnswerWwwExampleOrg(): creating answer for www.example.org" << endl;
+        }
+
+        // Give a response for www.example.org.
+        RRsetPtr www_example_org_a(new RRset(Name("www.example.org."), RRClass::IN(), RRType::A(), RRTTL(300)));
+        www_example_org_a->addRdata(createRdata(RRType::A(), RRClass::IN(), WWW_EXAMPLE_ORG));
+        msg.addRRset(Message::SECTION_ANSWER, www_example_org_a);
+
+        // ... and add the Authority and Additional sections. (These are the
+        // same as in the referral to example.org from the .org nameserver.)
+        setReferralExampleOrg(msg);
+    }
+
+    /// \brief UDP Receive Handler
+    ///
+    /// This is invoked when a message is received over UDP from the
+    /// RecursiveQuery object under test.  It formats an answer and sends it
+    /// asynchronously, with the UdpSendHandler method being specified as the
+    /// completion handler.
+    ///
+    /// \param ec ASIO error code, completion code of asynchronous I/O issued
+    ///        by the "server" to receive data.
+    /// \param length Amount of data received.
+    void udpReceiveHandler(error_code ec = error_code(), size_t length = 0) {
+        if (debug_) {
+            cout << "udpReceiveHandler(): error = " << ec.value() <<
+                    ", length = " << length << ", last state = " << last_ <<
+                    ", expected state = " << expected_ << endl;
+        }
+
+        // Expected state should be one greater than the last state.
+        EXPECT_EQ(static_cast<int>(expected_), static_cast<int>(last_) + 1);
+        last_ = expected_;
+
+        // The QID in the incoming data is random so set it to 0 for the
+        // data comparison check. (It is set to 0 in the buffer containing
+        // the expected data.)
+        uint16_t qid = readUint16(udp_receive_buffer_);
+        udp_receive_buffer_[0] = udp_receive_buffer_[1] = 0;
+
+        // Check that question we received is what was expected.
+        checkReceivedPacket(udp_receive_buffer_, length);
+
+        // The message returned depends on what state we are in.  Set up
+        // common stuff first: bits not mentioned are set to 0.
+        Message msg(Message::RENDER);
+        setCommonMessage(msg, qid);
+
+        // Set up state-dependent bits:
+        switch (expected_) {
+        case UDP_ROOT:
+            // Return a referral to org.  We then expect to query the "org"
+            // nameservers over UDP next.
+            setReferralOrg(msg);
+            expected_ = UDP_ORG;
+            break;
+
+         case UDP_ORG:
+            // Return a referral to example.org.  We explicitly set the TC bit to
+            // force a repeat query to the .org nameservers over TCP.
+            setReferralExampleOrg(msg);
+            if (debug_) {
+                cout << "udpReceiveHandler(): setting TC bit" << endl;
+            }
+            msg.setHeaderFlag(Message::HEADERFLAG_TC);
+            expected_ = TCP_ORG;
+            break;
+
+         case UDP_EXAMPLE_ORG:
+            // Return the answer to the question.
+            setAnswerWwwExampleOrg(msg);
+            expected_ = COMPLETE;
+            break;
+
+         default:
+            FAIL() << "UdpReceiveHandler called with unknown state";
+        }
+
+        // Convert to wire format
+        udp_send_buffer_->clear();
+        MessageRenderer renderer(*udp_send_buffer_);
+        msg.toWire(renderer);
+
+        // Return a message back to the IOFetch object (after setting the
+        // expected length of data for the check in the send handler).
+        udp_length_ = udp_send_buffer_->getLength();
+        udp_socket_.async_send_to(asio::buffer(udp_send_buffer_->getData(),
+                                               udp_send_buffer_->getLength()),
+                                  udp_remote_,
+                                  boost::bind(&RecursiveQueryTest2::udpSendHandler,
+                                              this, _1, _2));
+    }
+
+    /// \brief UDP Send Handler
+    ///
+    /// Called when a send operation of the UDP server (i.e. a response
+    /// being sent to the RecursiveQuery) has completed, this re-issues
+    /// a read call.
+    ///
+    /// \param ec Completion error code of the send.
+    /// \param length Actual number of bytes sent.
+    void udpSendHandler(error_code ec = error_code(), size_t length = 0) {
+        if (debug_) {
+            cout << "udpSendHandler(): error = " << ec.value() <<
+                    ", length = " << length << endl;
+        }
+
+        // Check send was OK
+        EXPECT_EQ(0, ec.value());
+        EXPECT_EQ(udp_length_, length);
+
+        // Reissue the receive call to await the next message.
+        udp_socket_.async_receive_from(
+            asio::buffer(udp_receive_buffer_, sizeof(udp_receive_buffer_)),
+            udp_remote_,
+            boost::bind(&RecursiveQueryTest2::udpReceiveHandler, this, _1, _2));
+    }
+
+    /// \brief Completion Handler for Accepting TCP Data
+    ///
+    /// Called when the remote system connects to the "TCP server".  It issues
+    /// an asynchronous read on the socket to read data.
+    ///
+    /// \param socket Socket on which data will be received
+    /// \param ec Boost error code, value should be zero.
+    void tcpAcceptHandler(error_code ec = error_code(), size_t length = 0) {
+        if (debug_) {
+            cout << "tcpAcceptHandler(): error = " << ec.value() <<
+                    ", length = " << length << endl;
+        }
+
+        // Expect that the accept completed without a problem.
+        EXPECT_EQ(0, ec.value());
+
+        // Initiate a read on the socket, indicating that nothing has yet been
+        // received.
+        tcp_cumulative_ = 0;
+        tcp_socket_.async_receive(
+            asio::buffer(tcp_receive_buffer_, sizeof(tcp_receive_buffer_)),
+            boost::bind(&RecursiveQueryTest2::tcpReceiveHandler, this, _1, _2));
+    }
+
+    /// \brief Completion Handler for Receiving TCP Data
+    ///
+    /// Reads data from the RecursiveQuery object and loops, reissuing reads,
+    /// until all the message has been read.  It then returns an appropriate
+    /// response.
+    ///
+    /// \param socket Socket to use to send the answer
+    /// \param ec ASIO error code, completion code of asynchronous I/O issued
+    ///        by the "server" to receive data.
+    /// \param length Amount of data received.
+    void tcpReceiveHandler(error_code ec = error_code(), size_t length = 0) {
+        if (debug_) {
+            cout << "tcpReceiveHandler(): error = " << ec.value() <<
+                    ", length = " << length <<
+                    ", cumulative = " << tcp_cumulative_ << endl;
+        }
+
+        // Expect that the receive completed without a problem.
+        EXPECT_EQ(0, ec.value());
+
+        // Have we received all the data?  We know this by checking if the two-
+        // byte length count in the message is equal to the data received.
+        tcp_cumulative_ += length;
+        bool complete = false;
+        if (tcp_cumulative_ > 2) {
+            uint16_t dns_length = readUint16(tcp_receive_buffer_);
+            complete = ((dns_length + 2) == tcp_cumulative_);
+        }
+
+        if (!complete) {
+            if (debug_) {
+                cout << "tcpReceiveHandler(): read not complete, " <<
+                        "issuing another read" << endl;
+            }
+
+            // Not complete yet, issue another read.
+            tcp_socket_.async_receive(
+                asio::buffer(tcp_receive_buffer_ + tcp_cumulative_,
+                             sizeof(tcp_receive_buffer_) - tcp_cumulative_),
+                boost::bind(&RecursiveQueryTest2::tcpReceiveHandler, this, _1, _2));
+            return;
+        }
+
+        // Have received a TCP message.  Expected state should be one greater
+        // than the last state.
+        EXPECT_EQ(static_cast<int>(expected_), static_cast<int>(last_) + 1);
+        last_ = expected_;
+
+        // Check that question we received is what was expected.  Note that we
+        // have to ignore the two-byte header in order to parse the message.
+        checkReceivedPacket(tcp_receive_buffer_ + 2, length - 2);
+
+        // Return a message back.  This is a referral to example.org, which
+        // should result in another query over UDP.  Note the setting of the
+        // QID in the returned message with what was in the received message.
+        Message msg(Message::RENDER);
+        setCommonMessage(msg, readUint16(tcp_receive_buffer_));
+        setReferralExampleOrg(msg);
+
+        // Convert to wire format
+        tcp_send_buffer_->clear();
+        MessageRenderer renderer(*tcp_send_buffer_);
+        msg.toWire(renderer);
+
+        // Expected next state (when checked) is the UDP query to example.org.
+        // Also, take this opportunity to clear the accumulated read count in
+        // readiness for the next read. (If any - at present, there is only
+        // one read in the test, although extensions to this test suite could
+        // change that.)
+        expected_ = UDP_EXAMPLE_ORG;
+        tcp_cumulative_ = 0;
+
+        // We'll write the message in two parts, the count and the message
+        // itself. This saves having to prepend the count onto the start of a
+        // buffer.  When specifying the send handler, the expected size of the
+        // data written is passed as the first parameter so that the handler
+        // can check it.
+        uint8_t count[2];
+        writeUint16(tcp_send_buffer_->getLength(), count);
+        tcp_socket_.async_send(asio::buffer(count, 2),
+                               boost::bind(&RecursiveQueryTest2::tcpSendHandler, this,
+                                           2, _1, _2));
+        tcp_socket_.async_send(asio::buffer(tcp_send_buffer_->getData(),
+                                            tcp_send_buffer_->getLength()),
+                               boost::bind(&RecursiveQueryTest2::tcpSendHandler, this,
+                                           tcp_send_buffer_->getLength(), _1, _2));
+    }
+
+    /// \brief Completion Handler for Sending TCP data
+    ///
+    /// Called when the asynchronous send of data back to the RecursiveQuery
+    /// by the TCP "server" in this class has completed.  (This send has to
+    /// be asynchronous because control needs to return to the caller in order
+    /// for the IOService "run()" method to be called to run the handlers.)
+    ///
+    /// \param expected_length Number of bytes that were expected to have been sent.
+    /// \param ec Boost error code, value should be zero.
+    /// \param length Number of bytes sent.
+    void tcpSendHandler(size_t expected_length = 0, error_code ec = error_code(),
+                        size_t length = 0)
+    {
+        if (debug_) {
+            cout << "tcpSendHandler(): error = " << ec.value() <<
+                    ", length = " << length <<
+                    ", (expected length = " << expected_length << ")" << endl;
+        }
+        EXPECT_EQ(0, ec.value());       // Expect no error
+        EXPECT_EQ(expected_length, length);    // And that amount sent is as expected
+    }
+
+    /// \brief Check Received Packet
+    ///
+    /// Checks the packet received from the RecursiveQuery object to ensure
+    /// that the question is what is expected.
+    ///
+    /// \param data Start of data.  This is the start of the received buffer in
+    ///        the case of UDP data, and an offset into the buffer past the
+    ///        count field for TCP data.
+    /// \param length Length of data.
+    void checkReceivedPacket(uint8_t* data, size_t length) {
+
+        // Decode the received buffer.
+        InputBuffer buffer(data, length);
+        Message message(Message::PARSE);
+        message.fromWire(buffer);
+
+        // Check the packet.
+        EXPECT_FALSE(message.getHeaderFlag(Message::HEADERFLAG_QR));
+
+        Question question = **(message.beginQuestion());
+        EXPECT_TRUE(question == *question_);
+    }
+};
+
+/// \brief Resolver Callback Object
+///
+/// Holds the success and failure callback methods for the resolver
+class ResolverCallback : public isc::resolve::ResolverInterface::Callback {
+public:
+    /// \brief Constructor
+    ResolverCallback(IOService& service) :
+        service_(service), run_(false), status_(false), debug_(DEBUG_PRINT)
+    {}
+
+    /// \brief Destructor
+    virtual ~ResolverCallback()
+    {}
+
+    /// \brief Resolver Callback Success
+    ///
+    /// Called if the resolver detects that the call has succeeded.
+    ///
+    /// \param response Answer to the question.
+    virtual void success(const isc::dns::MessagePtr response) {
+        if (debug_) {
+            cout << "ResolverCallback::success(): answer received" << endl;
+        }
+
+        // There should be one RR each  in the question and answer sections, and
+        // two RRs in each of the the authority and additional sections.
+        EXPECT_EQ(1, response->getRRCount(Message::SECTION_QUESTION));
+        EXPECT_EQ(1, response->getRRCount(Message::SECTION_ANSWER));
+        EXPECT_EQ(2, response->getRRCount(Message::SECTION_AUTHORITY));
+        EXPECT_EQ(2, response->getRRCount(Message::SECTION_ADDITIONAL));
+
+        // Check the answer - that the RRset is there...
+        EXPECT_TRUE(response->hasRRset(Message::SECTION_ANSWER,
+                                       RRsetPtr(new RRset(Name("www.example.org."),
+                                                RRClass::IN(),
+                                                RRType::A(),
+                                                RRTTL(300)))));
+        const RRsetIterator rrset_i = response->beginSection(Message::SECTION_ANSWER);
+
+        // ... get iterator into the Rdata of this RRset and point to first
+        // element...
+        const RdataIteratorPtr rdata_i = (*rrset_i)->getRdataIterator();
+        rdata_i->first();
+
+        // ... and check it is what we expect.
+        EXPECT_EQ(string(WWW_EXAMPLE_ORG), rdata_i->getCurrent().toText());
+
+        // Flag completion
+        run_ = true;
+        status_ = true;
+
+        service_.stop();    // Cause run() to exit.
+    }
+
+    /// \brief Resolver Failure Completion
+    ///
+    /// Called if the resolver detects that the resolution has failed.
+    virtual void failure() {
+        if (debug_) {
+            cout << "ResolverCallback::success(): resolution failure" << endl;
+        }
+        FAIL() << "Resolver reported completion failure";
+
+        // Flag completion
+        run_ = true;
+        status_ = false;
+
+        service_.stop();    // Cause run() to exit.
+    }
+
+    /// \brief Return status of "run" flag
+    bool getRun() const {
+        return (run_);
+    }
+
+    /// \brief Return "status" flag
+    bool getStatus() const {
+        return (status_);
+    }
+
+private:
+    IOService&      service_;       ///< Service handling the run queue
+    bool            run_;           ///< Set true when completion handler run
+    bool            status_;        ///< Set true for success, false on error
+    bool            debug_;         ///< Debug flag
+};
+
+// Sets up the UDP and TCP "servers", then tries a resolution.
+
+TEST_F(RecursiveQueryTest2, Resolve) {
+
+    // Set up the UDP server and issue the first read.  The endpoint from which
+    // the query is sent is put in udp_endpoint_ when the read completes, which
+    // is referenced in the callback as the place to which the response is sent.
+    udp_socket_.set_option(socket_base::reuse_address(true));
+    udp_socket_.bind(udp::endpoint(address::from_string(TEST_ADDRESS), TEST_PORT));
+    udp_socket_.async_receive_from(asio::buffer(udp_receive_buffer_,
+                                                sizeof(udp_receive_buffer_)),
+                                   udp_remote_,
+                                   boost::bind(&RecursiveQueryTest2::udpReceiveHandler,
+                                               this, _1, _2));
+
+    // Set up the TCP server and issue the accept.  Acceptance will cause the
+    // read to be issued.
+    tcp::acceptor acceptor(service_.get_io_service(),
+                           tcp::endpoint(tcp::v4(), TEST_PORT));
+    acceptor.async_accept(tcp_socket_,
+                          boost::bind(&RecursiveQueryTest2::tcpAcceptHandler,
+                                      this, _1, 0));
+
+    // Set up the RecursiveQuery object.
+    std::vector<std::pair<std::string, uint16_t> > upstream;         // Empty
+    std::vector<std::pair<std::string, uint16_t> > upstream_root;    // Empty
+    RecursiveQuery query(dns_service_, *nsas_, cache_,
+                         upstream, upstream_root);
+    query.setTestServer(TEST_ADDRESS, TEST_PORT);
+
+    // Set up callback for the tor eceive notification that the query has
+    // completed.
+    isc::resolve::ResolverInterface::CallbackPtr
+        resolver_callback(new ResolverCallback(service_));
+
+    // Kick off the resolution process.  We expect the first question to go to
+    // "root".
+    expected_ = UDP_ROOT;
+    query.resolve(question_, resolver_callback);
+    service_.run();
+
+    // Check what ran. (We have to cast the callback to ResolverCallback as we
+    // lost the information on the derived class when we used a
+    // ResolverInterface::CallbackPtr to store a pointer to it.)
+    ResolverCallback* rc = static_cast<ResolverCallback*>(resolver_callback.get());
+    EXPECT_TRUE(rc->getRun());
+    EXPECT_TRUE(rc->getStatus());
+}
+
+} // namespace asiolink

src/lib/resolve/tests/response_classifier_unittest.cc

@@ -80,6 +80,8 @@ public:
             RRType::CNAME(), RRTTL(300))),
         rrs_in_ns_(new RRset(Name("example.com"), RRClass::IN(),
             RRType::NS(), RRTTL(300))),
+        rrs_in_soa_(new RRset(Name("example.com"), RRClass::IN(),
+            RRType::SOA(), RRTTL(300))),
         rrs_in_txt_www(new RRset(Name("www.example.com"), RRClass::IN(),
             RRType::TXT(), RRTTL(300))),
         cname_target("."),
@@ -115,6 +117,9 @@ public:
         // Set up an imaginary NS RRset for an authority section
         rrs_in_ns_->addRdata(ConstRdataPtr(new NS(Name("ns0.isc.org"))));
         rrs_in_ns_->addRdata(ConstRdataPtr(new NS(Name("ns0.example.org"))));
+        
+        // And an imaginary SOA
+        rrs_in_soa_->addRdata(ConstRdataPtr(new SOA(Name("ns0.example.org"), Name("root.example.org"), 1, 2, 3, 4, 5)));
 
         // Set up the records for the www host
         rrs_in_a_www->addRdata(ConstRdataPtr(new A("1.2.3.4")));
@@ -146,6 +151,7 @@ public:
     RRsetPtr    rrs_in_cname_www1;  // www1.example.com IN CNAME
     RRsetPtr    rrs_in_cname_www2;  // www2.example.com IN CNAME
     RRsetPtr    rrs_in_ns_;         // example.com IN NS
+    RRsetPtr    rrs_in_soa_;        // example.com IN SOA
     RRsetPtr    rrs_in_txt_www;     // www.example.com IN TXT
     Name        cname_target;       // Used in response classifier to
                                     // store the target of a possible
@@ -349,6 +355,17 @@ TEST_F(ResponseClassifierTest, EmptyAnswerReferral) {
 
 }
 
+// Test if we get a NOERROR answer that contains neither an actual
+// answer nor a delegation
+TEST_F(ResponseClassifierTest, NoErrorNoData) {
+
+    msg_a.addRRset(Message::SECTION_AUTHORITY, rrs_in_soa_);
+    EXPECT_EQ(ResponseClassifier::NXRRSET,
+        ResponseClassifier::classify(qu_in_a_www, msg_a, cname_target,
+                                     cname_count));
+
+}
+
 // Check the case where we have a simple answer in the answer section.  This
 // occurs when the QNAME/QTYPE/QCLASS matches one of the RRsets in the
 // answer section - expect when the QTYPE is ANY, in which case the match

src/lib/server_common/tests/Makefile.am

@@ -33,6 +33,10 @@ run_unittests_LDFLAGS = $(AM_LDFLAGS) $(GTEST_LDFLAGS)
 run_unittests_LDADD = $(GTEST_LDADD)
 
 run_unittests_LDADD += $(top_builddir)/src/lib/server_common/libserver_common.la
+run_unittests_LDADD += $(top_builddir)/src/lib/exceptions/libexceptions.la
+run_unittests_LDADD += $(top_builddir)/src/lib/asiolink/libasiolink.la
+run_unittests_LDADD += $(top_builddir)/src/lib/cc/libcc.la
+run_unittests_LDADD += $(top_builddir)/src/lib/dns/libdns++.la
 endif
 
 noinst_PROGRAMS = $(TESTS)

src/lib/testutils/Makefile.am

@@ -11,6 +11,7 @@ libtestutils_la_SOURCES = srv_test.h srv_test.cc
 libtestutils_la_SOURCES += dnsmessage_test.h dnsmessage_test.cc
 libtestutils_la_SOURCES += mockups.h
 libtestutils_la_CPPFLAGS = $(AM_CPPFLAGS) $(GTEST_INCLUDES)
+libtestutils_la_LIBADD = $(top_builddir)/src/lib/asiolink/libasiolink.la
 endif
 
 EXTRA_DIST = portconfig.h

tests/system/Makefile.am

@@ -7,7 +7,10 @@ distclean-local:
 # Most of the files under this directory (including test subdirectories)
 # must be listed in EXTRA_DIST.
 EXTRA_DIST = README cleanall.sh ifconfig.sh start.pl stop.pl run.sh runall.sh
+EXTRA_DIST += common/default_user.csv
 EXTRA_DIST += glue/auth.good glue/example.good glue/noglue.good glue/test.good
 EXTRA_DIST += glue/tests.sh glue/clean.sh
 EXTRA_DIST += glue/nsx1/com.db glue/nsx1/net.db glue/nsx1/root-servers.nil.db
 EXTRA_DIST += glue/nsx1/root.db
+EXTRA_DIST += bindctl/tests.sh bindctl/clean.sh bindctl/setup.sh
+EXTRA_DIST += bindctl/nsx1/root.db bindctl/nsx1/example-normalized.db

tests/system/README

@@ -18,6 +18,7 @@ set to point to the top directory of the source tree.
 There are multiple test suites, each in a separate subdirectory and
 involving a different DNS setup.  They are:
 
+  bindctl/      Some basic management operations using the bindctl tool
   glue/		Glue handling tests
 (the following tests are planned to be added soon)
   dnssec/	DNSSEC tests

tests/system/bindctl/clean.sh

@@ -0,0 +1,20 @@
+#!/bin/sh
+#
+# Copyright (C) 2004, 2007  Internet Systems Consortium, Inc. ("ISC")
+# Copyright (C) 2000, 2001  Internet Software Consortium.
+#
+# Permission to use, copy, modify, and/or distribute this software for any
+# purpose with or without fee is hereby granted, provided that the above
+# copyright notice and this permission notice appear in all copies.
+#
+# THE SOFTWARE IS PROVIDED "AS IS" AND ISC DISCLAIMS ALL WARRANTIES WITH
+# REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
+# AND FITNESS.  IN NO EVENT SHALL ISC BE LIABLE FOR ANY SPECIAL, DIRECT,
+# INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
+# LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
+# OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
+# PERFORMANCE OF THIS SOFTWARE.
+
+rm -f */b10-config.db
+rm -f dig.out.* bindctl.out.*
+rm -f */msgq_socket */zone.sqlite3

tests/system/bindctl/nsx1/b10-config.db.template.in

@@ -0,0 +1,10 @@
+{"version": 2,
+ "Auth": {
+   "listen_on": [{"address": "10.53.0.1", "port": 53210}],
+   "database_file": "@abs_builddir@/zone.sqlite3",
+   "statistics-interval": 1
+ },
+ "Xfrout": {
+   "log_file": "@abs_builddir@/Xfrout.log"
+ }
+}

tests/system/bindctl/nsx1/example-normalized.db

@@ -0,0 +1,3 @@
+com.					      300 IN SOA	postmaster.example. ns.example.com. 2000042100 600 600 1200 600
+com.					      300 IN NS		ns.example.com.
+ns.example.com.				      300 IN A		192.0.2.2

tests/system/bindctl/nsx1/root.db

@@ -0,0 +1,25 @@
+; Copyright (C) 2004, 2007  Internet Systems Consortium, Inc. ("ISC")
+; Copyright (C) 2000, 2001  Internet Software Consortium.
+;
+; Permission to use, copy, modify, and/or distribute this software for any
+; purpose with or without fee is hereby granted, provided that the above
+; copyright notice and this permission notice appear in all copies.
+;
+; THE SOFTWARE IS PROVIDED "AS IS" AND ISC DISCLAIMS ALL WARRANTIES WITH
+; REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
+; AND FITNESS.  IN NO EVENT SHALL ISC BE LIABLE FOR ANY SPECIAL, DIRECT,
+; INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
+; LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
+; OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
+; PERFORMANCE OF THIS SOFTWARE.
+
+$TTL 300
+. 			IN SOA	postmaster.example. a.root.servers.nil. (
+				2000042100   	; serial
+				600         	; refresh
+				600         	; retry
+				1200    	; expire
+				600       	; minimum
+				)
+.			NS	ns.example.com.
+ns.example.com.		A	192.0.2.1

tests/system/bindctl/setup.sh

@@ -0,0 +1,26 @@
+#!/bin/sh
+#
+# Copyright (C) 2011  Internet Systems Consortium, Inc. ("ISC")
+#
+# Permission to use, copy, modify, and/or distribute this software for any
+# purpose with or without fee is hereby granted, provided that the above
+# copyright notice and this permission notice appear in all copies.
+#
+# THE SOFTWARE IS PROVIDED "AS IS" AND ISC DISCLAIMS ALL WARRANTIES WITH
+# REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
+# AND FITNESS.  IN NO EVENT SHALL ISC BE LIABLE FOR ANY SPECIAL, DIRECT,
+# INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
+# LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
+# OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
+# PERFORMANCE OF THIS SOFTWARE.
+
+SYSTEMTESTTOP=..
+. $SYSTEMTESTTOP/conf.sh
+
+SUBTEST_TOP=${TEST_TOP}/bindctl
+
+cp ${SUBTEST_TOP}/nsx1/b10-config.db.template ${SUBTEST_TOP}/nsx1/b10-config.db
+
+rm -f ${SUBTEST_TOP}/*/zone.sqlite3
+${B10_LOADZONE} -o . -d ${SUBTEST_TOP}/nsx1/zone.sqlite3 \
+	${SUBTEST_TOP}//nsx1/root.db

tests/system/bindctl/tests.sh

@@ -0,0 +1,106 @@
+#!/bin/sh
+#
+# Copyright (C) 2011  Internet Systems Consortium, Inc. ("ISC")
+#
+# Permission to use, copy, modify, and/or distribute this software for any
+# purpose with or without fee is hereby granted, provided that the above
+# copyright notice and this permission notice appear in all copies.
+#
+# THE SOFTWARE IS PROVIDED "AS IS" AND ISC DISCLAIMS ALL WARRANTIES WITH
+# REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
+# AND FITNESS.  IN NO EVENT SHALL ISC BE LIABLE FOR ANY SPECIAL, DIRECT,
+# INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
+# LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
+# OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
+# PERFORMANCE OF THIS SOFTWARE.
+
+SYSTEMTESTTOP=..
+. $SYSTEMTESTTOP/conf.sh
+
+#
+# Do bindctl tests.
+#
+
+status=0
+n=0
+
+echo "I:Checking b10-auth is working by default ($n)"
+$DIG +norec @10.53.0.1 -p 53210 ns.example.com. A >dig.out.$n || status=1
+# perform a simple check on the output (digcomp would be too much for this)
+grep 192.0.2.1 dig.out.$n > /dev/null || status=1
+if [ $status != 0 ]; then echo "I:failed"; fi
+n=`expr $n + 1`
+
+echo "I:Checking BIND 10 statistics after a pose ($n)"
+# wait for 2sec to make sure b10-stats gets the latest statistics.
+# note that we set statistics-interval to 1.
+sleep 2
+echo 'Stats show
+' | $RUN_BINDCTL \
+	--csv-file-dir=$BINDCTL_CSV_DIR > bindctl.out.$n || status=1
+# the server should have received 1 UDP and 1 TCP queries (TCP query was
+# sent from the server startup script)
+grep "\"auth.queries.tcp\": 1," bindctl.out.$n > /dev/null || status=1
+grep "\"auth.queries.udp\": 1," bindctl.out.$n > /dev/null || status=1
+if [ $status != 0 ]; then echo "I:failed"; fi
+n=`expr $n + 1`
+
+echo "I:Stopping b10-auth and checking that ($n)"
+echo 'config set Boss/start_auth false
+config commit
+quit
+' | $RUN_BINDCTL \
+	--csv-file-dir=$BINDCTL_CSV_DIR 2>&1 > /dev/null || status=1
+# dig should exit with a failure code.
+$DIG +tcp +norec @10.53.0.1 -p 53210 ns.example.com. A && status=1
+if [ $status != 0 ]; then echo "I:failed"; fi
+n=`expr $n + 1`
+
+echo "I:Restarting b10-auth and checking that ($n)"
+echo 'config set Boss/start_auth true
+config commit
+quit
+' | $RUN_BINDCTL \
+	--csv-file-dir=$BINDCTL_CSV_DIR 2>&1 > /dev/null || status=1
+$DIG +norec @10.53.0.1 -p 53210 ns.example.com. A >dig.out.$n || status=1
+grep 192.0.2.1 dig.out.$n > /dev/null || status=1
+if [ $status != 0 ]; then echo "I:failed"; fi
+n=`expr $n + 1`
+
+echo "I:Rechecking BIND 10 statistics after a pose ($n)"
+sleep 2
+echo 'Stats show
+' | $RUN_BINDCTL \
+	--csv-file-dir=$BINDCTL_CSV_DIR > bindctl.out.$n || status=1
+# The statistics counters should have been reset while stop/start.
+grep "\"auth.queries.tcp\": 0," bindctl.out.$n > /dev/null || status=1
+grep "\"auth.queries.udp\": 1," bindctl.out.$n > /dev/null || status=1
+if [ $status != 0 ]; then echo "I:failed"; fi
+n=`expr $n + 1`
+
+echo "I:Changing the data source from sqlite3 to in-memory ($n)"
+DATASRC_SPEC='[{"type": "memory", "zones": [{"origin": "com","file":'
+DATASRC_SPEC="${DATASRC_SPEC} \"${TEST_TOP}/bindctl/nsx1/example-normalized.db\"}]}]"
+echo "config set Auth/datasources ${DATASRC_SPEC}
+config commit
+quit
+" | $RUN_BINDCTL \
+	--csv-file-dir=$BINDCTL_CSV_DIR > bindctl.out.$n || status=1
+$DIG +norec @10.53.0.1 -p 53210 ns.example.com. A >dig.out.$n || status=1
+grep 192.0.2.2 dig.out.$n > /dev/null || status=1
+if [ $status != 0 ]; then echo "I:failed"; fi
+n=`expr $n + 1`
+
+echo "I:Rechecking BIND 10 statistics after changing the datasource ($n)"
+sleep 2
+echo 'Stats show
+' | $RUN_BINDCTL \
+	--csv-file-dir=$BINDCTL_CSV_DIR > bindctl.out.$n || status=1
+# The statistics counters shouldn't be reset due to hot-swapping datasource.
+grep "\"auth.queries.tcp\": 0," bindctl.out.$n > /dev/null || status=1
+grep "\"auth.queries.udp\": 2," bindctl.out.$n > /dev/null || status=1
+if [ $status != 0 ]; then echo "I:failed"; fi
+n=`expr $n + 1`
+
+echo "I:exit status: $status"
+exit $status

tests/system/common/default_user.csv

@@ -0,0 +1 @@
+root,bind10

tests/system/conf.sh.in

@@ -34,10 +34,13 @@ if [ -z $BIND9_TOP ]; then
 	exit 1
 fi
 
-# Find the top of the source tree.
+# Find the top of the source and test trees.
 TOP=@abs_top_srcdir@
+TEST_TOP=@abs_builddir@
 
 RUN_BIND10=$TOP/src/bin/bind10/run_bind10.sh
+RUN_BINDCTL=$TOP/src/bin/bindctl/run_bindctl.sh
+BINDCTL_CSV_DIR=@abs_srcdir@/common/
 B10_LOADZONE=$TOP/src/bin/loadzone/run_loadzone.sh
 BIND9_NAMED=$BIND9_TOP/bin/named/named
 DIG=$BIND9_TOP/bin/dig/dig
@@ -45,8 +48,8 @@ DIG=$BIND9_TOP/bin/dig/dig
 TESTSOCK=$BIND9_TOP/bin/tests/system/testsock.pl
 DIGCOMP=$BIND9_TOP/bin/tests/system/digcomp.pl
 
-SUBDIRS="glue"
-#SUBDIRS="dnssec glue masterfile xfer"
+SUBDIRS="bindctl glue"
+#SUBDIRS="dnssec masterfile xfer"
 
 # PERL will be an empty string if no perl interpreter was found.
 PERL=@PERL@