Problem: test failure on CI due to ZMQ_EVENT_HANDSHAKE_FAILED_NO_DETAIL/EPIPE problem

Solution: add workaround at another place, unify two code fragments to remove duplication
This commit is contained in:
Simon Giesecke 2017-08-04 16:05:20 +02:00
parent aacb219acd
commit 7ba70e95e5

View File

@ -272,6 +272,48 @@ void expect_new_client_curve_bounce_fail (void *ctx,
close_zero_linger (client);
}
#ifdef ZMQ_BUILD_DRAFT_API
// expects that one or more occurrences of the expected event are received
// via the specified socket monitor
// returns the number of occurrences of the expected event
// interrupts, if a ZMQ_EVENT_HANDSHAKE_FAILED_NO_DETAIL/EPIPE occurs;
// in this case, 0 is returned
// this should be investigated further, see
// https://github.com/zeromq/libzmq/issues/2644
int expect_monitor_event_multiple (void *server_mon,
int expected_event,
int expected_err = -1)
{
int count_of_expected_events = 0;
int client_closed_connection = 0;
int timeout = -1;
int event;
int err;
while (
(event = get_monitor_event_with_timeout (server_mon, &err, NULL, timeout))
!= -1) {
timeout = 250;
// ignore errors with EPIPE, which happen sporadically, see above
if (event == ZMQ_EVENT_HANDSHAKE_FAILED_NO_DETAIL && err == EPIPE) {
fprintf (stderr, "Ignored event: %x (err = %i)\n", event, err);
client_closed_connection = 1;
break;
}
if (event != expected_event
|| (-1 != expected_err && err != expected_err)) {
fprintf (stderr, "Unexpected event: %x (err = %i)\n", event, err);
assert (false);
}
++count_of_expected_events;
}
assert (count_of_expected_events > 0 || client_closed_connection);
return count_of_expected_events;
}
#endif
void test_garbage_key(void *ctx,
void *server,
void *server_mon,
@ -284,67 +326,19 @@ void test_garbage_key(void *ctx,
client_secret, my_endpoint, server);
#ifdef ZMQ_BUILD_DRAFT_API
int timeout = -1;
int handshake_failed_encryption_event_count =
expect_monitor_event_multiple (server_mon,
ZMQ_EVENT_HANDSHAKE_FAILED_ENCRYPTION);
int handshake_failed_encryption_event_count = 0;
int handshake_failed_client_closed = 0;
int err;
int event;
int event_count = 0;
while (
(event = get_monitor_event_with_timeout (server_mon, &err, NULL, timeout))
!= -1) {
++event_count;
timeout = 250;
switch (event) {
case ZMQ_EVENT_HANDSHAKE_FAILED_ENCRYPTION:
++handshake_failed_encryption_event_count;
break;
case ZMQ_EVENT_HANDSHAKE_FAILED_NO_DETAIL:
// ignore errors with EPIPE, which happen sporadically
if (err == EPIPE) {
fprintf (stderr, "Ignored event: %x (err = %i)\n", event,
err);
++handshake_failed_client_closed;
continue;
}
default:
fprintf (stderr, "Unexpected event: %x (err = %i)\n", event,
err);
assert (false);
}
if (handshake_failed_encryption_event_count == 2
|| handshake_failed_client_closed == 1)
break;
}
fprintf (stderr,
"event_count == %i, "
"handshake_failed_encryption_event_count == %i, "
"handshake_failed_client_closed = %i\n",
event_count, handshake_failed_encryption_event_count,
handshake_failed_client_closed);
// handshake_failed_encryption_event_count should be two because
// handshake_failed_encryption_event_count should be at least two because
// expect_bounce_fail involves two exchanges
// however, with valgrind we see only one event (maybe the next one takes
// very long, or does not happen at all because something else takes very
// long)
// cases where handshake_failed_client_closed == 1 should be
// investigated further, see https://github.com/zeromq/libzmq/issues/2644
assert (handshake_failed_encryption_event_count >= 1
|| handshake_failed_client_closed == 1);
// Even though the client socket is closed, the server still handles HELLO
// messages. Output them for diagnostic purposes.
do {
int err;
event =
get_monitor_event_with_timeout (server_mon, &err, NULL, timeout);
if (event != -1) {
fprintf (stderr, "Flushed event: %x (errno = %i)\n", event, err);
}
} while (event != -1);
fprintf (stderr,
"count of ZMQ_EVENT_HANDSHAKE_FAILED_ENCRYPTION events: %i\n",
handshake_failed_encryption_event_count);
#endif
}
@ -560,7 +554,6 @@ void test_curve_security_zap_unsuccessful (void *ctx,
char *my_endpoint,
void *server,
void *server_mon,
int timeout,
int expected_event,
int expected_err)
{
@ -569,38 +562,23 @@ void test_curve_security_zap_unsuccessful (void *ctx,
my_endpoint, server);
#ifdef ZMQ_BUILD_DRAFT_API
int count_of_expected_events = 0;
int event;
int err;
while (
(event = get_monitor_event_with_timeout (server_mon, &err, NULL, timeout))
!= -1) {
if (event == expected_event) {
++count_of_expected_events;
if (err != expected_err) {
fprintf (stderr, "Unexpected event: %x (err = %i)\n", event,
err);
assert (false);
}
}
}
assert (count_of_expected_events > 0);
expect_monitor_event_multiple (server_mon, expected_event, expected_err);
#endif
// there may be more than one ZAP request due to repeated attempts by the client
assert (1 <= zmq_atomic_counter_value (zap_requests_handled));
}
void test_curve_security_zap_protocol_error(
void *ctx, char *my_endpoint, void *server, void *server_mon, int timeout)
void test_curve_security_zap_protocol_error (void *ctx,
char *my_endpoint,
void *server,
void *server_mon)
{
test_curve_security_zap_unsuccessful (
ctx, my_endpoint, server, server_mon, timeout,
test_curve_security_zap_unsuccessful (ctx, my_endpoint, server, server_mon,
#ifdef ZMQ_BUILD_DRAFT_API
ZMQ_EVENT_HANDSHAKE_FAILED_ZAP, EPROTO
ZMQ_EVENT_HANDSHAKE_FAILED_ZAP, EPROTO
#else
0, 0
0, 0
#endif
);
}
@ -712,7 +690,7 @@ int main (void)
&server_mon, my_endpoint,
&zap_handler_wrong_version);
test_curve_security_zap_protocol_error (ctx, my_endpoint, server,
server_mon, timeout);
server_mon);
shutdown_context_and_server_side (ctx, zap_thread, server, server_mon);
// wrong request id
@ -720,7 +698,7 @@ int main (void)
&server_mon, my_endpoint,
&zap_handler_wrong_request_id);
test_curve_security_zap_protocol_error (ctx, my_endpoint, server,
server_mon, timeout);
server_mon);
shutdown_context_and_server_side (ctx, zap_thread, server, server_mon);
// status invalid (not a 3-digit number)
@ -728,7 +706,7 @@ int main (void)
&server_mon, my_endpoint,
&zap_handler_wrong_status_invalid);
test_curve_security_zap_protocol_error (ctx, my_endpoint, server,
server_mon, timeout);
server_mon);
shutdown_context_and_server_side (ctx, zap_thread, server, server_mon);
// too many parts
@ -736,7 +714,7 @@ int main (void)
&server_mon, my_endpoint,
&zap_handler_too_many_parts);
test_curve_security_zap_protocol_error (ctx, my_endpoint, server,
server_mon, timeout);
server_mon);
shutdown_context_and_server_side (ctx, zap_thread, server, server_mon);
// ZAP non-standard cases
@ -748,12 +726,12 @@ int main (void)
// TODO is this usable? EAGAIN does not appear to be an appropriate error
// code, and the status text is completely lost
test_curve_security_zap_unsuccessful (
ctx, my_endpoint, server, server_mon, timeout,
test_curve_security_zap_unsuccessful (ctx, my_endpoint, server, server_mon,
#ifdef ZMQ_BUILD_DRAFT_API
ZMQ_EVENT_HANDSHAKE_FAILED_NO_DETAIL, EAGAIN
ZMQ_EVENT_HANDSHAKE_FAILED_NO_DETAIL,
EAGAIN
#else
0, 0
0, 0
#endif
);
shutdown_context_and_server_side (ctx, zap_thread, server, server_mon);