2
0
mirror of https://gitlab.isc.org/isc-projects/bind9 synced 2025-08-23 18:49:54 +00:00
bind/lib/isc/tests/netmgr_test.c

2334 lines
58 KiB
C
Raw Normal View History

Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
/*
* Copyright (C) Internet Systems Consortium, Inc. ("ISC")
*
* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, you can obtain one at https://mozilla.org/MPL/2.0/.
*
* See the COPYRIGHT file distributed with this work for additional
* information regarding copyright ownership.
*/
#if HAVE_CMOCKA
#include <sched.h> /* IWYU pragma: keep */
#include <setjmp.h>
#include <signal.h>
#include <stdarg.h>
#include <stdlib.h>
#include <unistd.h>
#include <uv.h>
#define UNIT_TESTING
#include <cmocka.h>
#include <isc/hp.h>
#include <isc/nonce.h>
#include <isc/os.h>
#include <isc/quota.h>
#include <isc/refcount.h>
#include <isc/sockaddr.h>
#include <isc/thread.h>
#include <isc/util.h>
#include "uv_wrap.h"
#define KEEP_BEFORE
#include "../netmgr/netmgr-int.h"
#include "../netmgr/udp.c"
#include "../netmgr/uv-compat.c"
#include "../netmgr/uv-compat.h"
#include "isctest.h"
isc_nm_t *listen_nm = NULL;
isc_nm_t *connect_nm = NULL;
static isc_sockaddr_t udp_listen_addr;
static isc_sockaddr_t udp_connect_addr;
static isc_sockaddr_t tcp_listen_addr;
static isc_sockaddr_t tcp_connect_addr;
static isc_tlsctx_t *tcp_listen_tlsctx = NULL;
static isc_tlsctx_t *tcp_connect_tlsctx = NULL;
static uint64_t send_magic = 0;
static uint64_t stop_magic = 0;
static uv_buf_t send_msg = { .base = (char *)&send_magic,
.len = sizeof(send_magic) };
static uv_buf_t stop_msg = { .base = (char *)&stop_magic,
.len = sizeof(stop_magic) };
static atomic_bool do_send = ATOMIC_VAR_INIT(false);
static unsigned int workers = 0;
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
static atomic_int_fast64_t nsends;
static int_fast64_t esends; /* expected sends */
static atomic_int_fast64_t ssends;
static atomic_int_fast64_t sreads;
static atomic_int_fast64_t saccepts;
static atomic_int_fast64_t cconnects;
static atomic_int_fast64_t csends;
static atomic_int_fast64_t creads;
static isc_refcount_t active_cconnects;
static isc_refcount_t active_csends;
static isc_refcount_t active_creads;
static isc_refcount_t active_ssends;
static isc_refcount_t active_sreads;
static isc_quota_t listener_quota;
static atomic_bool check_listener_quota;
static bool skip_long_tests = false;
#define SKIP_IN_CI \
if (skip_long_tests) { \
skip(); \
return; \
}
#define NSENDS 100
/* Timeouts in miliseconds */
#define T_INIT 120 * 1000
#define T_IDLE 120 * 1000
#define T_KEEPALIVE 120 * 1000
#define T_ADVERTISED 120 * 1000
#define T_CONNECT 30 * 1000
/* Wait for 1 second (1000 * 1000 microseconds) */
#define WAIT_REPEATS 1000
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
#define T_WAIT 1000 /* In microseconds */
#define WAIT_FOR(v, op, val) \
{ \
X(v); \
int_fast64_t __r = WAIT_REPEATS; \
int_fast64_t __o = 0; \
do { \
int_fast64_t __l = atomic_load(&v); \
if (__l op val) { \
break; \
}; \
if (__o == __l) { \
__r--; \
} else { \
__r = WAIT_REPEATS; \
} \
__o = __l; \
usleep(T_WAIT); \
} while (__r > 0); \
X(v); \
P(__r); \
assert_true(atomic_load(&v) op val); \
}
#define WAIT_FOR_EQ(v, val) WAIT_FOR(v, ==, val)
#define WAIT_FOR_NE(v, val) WAIT_FOR(v, !=, val)
#define WAIT_FOR_LE(v, val) WAIT_FOR(v, <=, val)
#define WAIT_FOR_LT(v, val) WAIT_FOR(v, <, val)
#define WAIT_FOR_GE(v, val) WAIT_FOR(v, >=, val)
#define WAIT_FOR_GT(v, val) WAIT_FOR(v, >, val)
#define DONE() atomic_store(&do_send, false);
#define CHECK_RANGE_FULL(v) \
{ \
int __v = atomic_load(&v); \
assert_true(__v > 1); \
}
#define CHECK_RANGE_HALF(v) \
{ \
int __v = atomic_load(&v); \
assert_true(__v > 1); \
}
/* Enable this to print values while running tests */
#undef PRINT_DEBUG
#ifdef PRINT_DEBUG
#define X(v) \
fprintf(stderr, "%s:%s:%d:%s = %" PRId64 "\n", __func__, __FILE__, \
__LINE__, #v, atomic_load(&v))
#define P(v) fprintf(stderr, #v " = %" PRId64 "\n", v)
#define F() \
fprintf(stderr, "%s(%p, %s, %p)\n", __func__, handle, \
isc_result_totext(eresult), cbarg)
#else
#define X(v)
#define P(v)
#define F()
#endif
#define atomic_assert_int_eq(val, exp) assert_int_equal(atomic_load(&val), exp)
#define atomic_assert_int_ne(val, exp) \
assert_int_not_equal(atomic_load(&val), exp)
#define atomic_assert_int_le(val, exp) assert_true(atomic_load(&val) <= exp)
#define atomic_assert_int_lt(val, exp) assert_true(atomic_load(&val) > exp)
#define atomic_assert_int_ge(val, exp) assert_true(atomic_load(&val) >= exp)
#define atomic_assert_int_gt(val, exp) assert_true(atomic_load(&val) > exp)
static int
_setup(void **state __attribute__((unused))) {
char *p = NULL;
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
if (workers == 0) {
workers = isc_os_ncpus();
}
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
p = getenv("ISC_TASK_WORKERS");
if (p != NULL) {
workers = atoi(p);
}
INSIST(workers != 0);
if (isc_test_begin(NULL, false, workers) != ISC_R_SUCCESS) {
return (-1);
}
isc_hp_init(4 * workers);
signal(SIGPIPE, SIG_IGN);
if (getenv("CI") == NULL || getenv("CI_ENABLE_ALL_TESTS") != NULL) {
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
esends = NSENDS * workers;
} else {
esends = workers;
skip_long_tests = true;
}
if (isc_tlsctx_createserver(NULL, NULL, &tcp_listen_tlsctx) !=
ISC_R_SUCCESS) {
return (-1);
}
if (isc_tlsctx_createclient(&tcp_connect_tlsctx) != ISC_R_SUCCESS) {
return (-1);
}
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
return (0);
}
static int
_teardown(void **state __attribute__((unused))) {
isc_tlsctx_free(&tcp_connect_tlsctx);
isc_tlsctx_free(&tcp_listen_tlsctx);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
isc_test_end();
return (0);
}
static int
setup_ephemeral_port(isc_sockaddr_t *addr, sa_family_t family) {
socklen_t addrlen = sizeof(*addr);
uv_os_sock_t fd;
int r;
isc_sockaddr_fromin6(addr, &in6addr_loopback, 0);
fd = socket(AF_INET6, family, 0);
if (fd < 0) {
perror("setup_ephemeral_port: socket()");
return (-1);
}
r = bind(fd, (const struct sockaddr *)&addr->type.sa,
sizeof(addr->type.sin6));
if (r != 0) {
perror("setup_ephemeral_port: bind()");
isc__nm_closesocket(fd);
return (r);
}
r = getsockname(fd, (struct sockaddr *)&addr->type.sa, &addrlen);
if (r != 0) {
perror("setup_ephemeral_port: getsockname()");
isc__nm_closesocket(fd);
return (r);
}
#if IPV6_RECVERR
#define setsockopt_on(socket, level, name) \
setsockopt(socket, level, name, &(int){ 1 }, sizeof(int))
r = setsockopt_on(fd, IPPROTO_IPV6, IPV6_RECVERR);
if (r != 0) {
perror("setup_ephemeral_port");
isc__nm_closesocket(fd);
return (r);
}
#endif
return (fd);
}
static int
nm_setup(void **state __attribute__((unused))) {
uv_os_sock_t tcp_listen_sock = -1;
uv_os_sock_t udp_listen_sock = -1;
udp_connect_addr = (isc_sockaddr_t){ .length = 0 };
isc_sockaddr_fromin6(&udp_connect_addr, &in6addr_loopback, 0);
udp_listen_addr = (isc_sockaddr_t){ .length = 0 };
udp_listen_sock = setup_ephemeral_port(&udp_listen_addr, SOCK_DGRAM);
if (udp_listen_sock < 0) {
return (-1);
}
isc__nm_closesocket(udp_listen_sock);
udp_listen_sock = -1;
tcp_connect_addr = (isc_sockaddr_t){ .length = 0 };
isc_sockaddr_fromin6(&tcp_connect_addr, &in6addr_loopback, 0);
tcp_listen_addr = (isc_sockaddr_t){ .length = 0 };
tcp_listen_sock = setup_ephemeral_port(&tcp_listen_addr, SOCK_STREAM);
if (tcp_listen_sock < 0) {
return (-1);
}
isc__nm_closesocket(tcp_listen_sock);
tcp_listen_sock = -1;
atomic_store(&do_send, true);
atomic_store(&nsends, esends);
atomic_store(&saccepts, 0);
atomic_store(&sreads, 0);
atomic_store(&ssends, 0);
atomic_store(&cconnects, 0);
atomic_store(&csends, 0);
atomic_store(&creads, 0);
isc_refcount_init(&active_cconnects, 0);
isc_refcount_init(&active_csends, 0);
isc_refcount_init(&active_creads, 0);
isc_refcount_init(&active_ssends, 0);
isc_refcount_init(&active_sreads, 0);
isc_nonce_buf(&send_magic, sizeof(send_magic));
isc_nonce_buf(&stop_magic, sizeof(stop_magic));
if (send_magic == stop_magic) {
return (-1);
}
listen_nm = isc_nm_start(test_mctx, workers);
assert_non_null(listen_nm);
isc_nm_settimeouts(listen_nm, T_INIT, T_IDLE, T_KEEPALIVE,
T_ADVERTISED);
connect_nm = isc_nm_start(test_mctx, workers);
assert_non_null(connect_nm);
isc_nm_settimeouts(connect_nm, T_INIT, T_IDLE, T_KEEPALIVE,
T_ADVERTISED);
isc_quota_init(&listener_quota, 0);
atomic_store(&check_listener_quota, false);
return (0);
}
static int
nm_teardown(void **state __attribute__((unused))) {
UNUSED(state);
isc_nm_destroy(&connect_nm);
assert_null(connect_nm);
isc_nm_destroy(&listen_nm);
assert_null(listen_nm);
WAIT_FOR_EQ(active_cconnects, 0);
WAIT_FOR_EQ(active_csends, 0);
WAIT_FOR_EQ(active_csends, 0);
WAIT_FOR_EQ(active_ssends, 0);
WAIT_FOR_EQ(active_sreads, 0);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
isc_refcount_destroy(&active_cconnects);
isc_refcount_destroy(&active_csends);
isc_refcount_destroy(&active_creads);
isc_refcount_destroy(&active_ssends);
isc_refcount_destroy(&active_sreads);
return (0);
}
/* Callbacks */
static void
noop_recv_cb(isc_nmhandle_t *handle, isc_result_t eresult, isc_region_t *region,
void *cbarg) {
UNUSED(handle);
UNUSED(eresult);
UNUSED(region);
UNUSED(cbarg);
}
static unsigned int
noop_accept_cb(isc_nmhandle_t *handle, unsigned int result, void *cbarg) {
UNUSED(handle);
UNUSED(result);
UNUSED(cbarg);
return (0);
}
static void
noop_connect_cb(isc_nmhandle_t *handle, isc_result_t result, void *cbarg) {
UNUSED(handle);
UNUSED(result);
UNUSED(cbarg);
isc_refcount_decrement(&active_cconnects);
}
static void
connect_send_cb(isc_nmhandle_t *handle, isc_result_t eresult, void *cbarg);
static void
connect_send(isc_nmhandle_t *handle);
static void
connect_send_cb(isc_nmhandle_t *handle, isc_result_t eresult, void *cbarg) {
isc_nmhandle_t *sendhandle = handle;
assert_non_null(sendhandle);
UNUSED(cbarg);
F();
if (eresult != ISC_R_SUCCESS) {
/* Send failed, we need to stop reading too */
isc_nm_cancelread(handle);
goto unref;
}
atomic_fetch_add(&csends, 1);
unref:
isc_refcount_decrement(&active_csends);
isc_nmhandle_detach(&sendhandle);
}
static void
connect_send(isc_nmhandle_t *handle) {
isc_nmhandle_t *sendhandle = NULL;
isc_refcount_increment0(&active_csends);
isc_nmhandle_attach(handle, &sendhandle);
if (atomic_fetch_sub(&nsends, 1) > 1) {
isc_nm_send(sendhandle, (isc_region_t *)&send_msg,
connect_send_cb, NULL);
} else {
isc_nm_send(sendhandle, (isc_region_t *)&stop_msg,
connect_send_cb, NULL);
}
}
static void
connect_read_cb(isc_nmhandle_t *handle, isc_result_t eresult,
isc_region_t *region, void *cbarg) {
uint64_t magic = 0;
UNUSED(cbarg);
assert_non_null(handle);
F();
if (eresult != ISC_R_SUCCESS) {
goto unref;
}
assert_int_equal(region->length, sizeof(magic));
atomic_fetch_add(&creads, 1);
memmove(&magic, region->base, sizeof(magic));
assert_true(magic == stop_magic || magic == send_magic);
unref:
atomic_fetch_sub(&active_creads, 1);
isc_nmhandle_detach(&handle);
}
static void
connect_connect_cb(isc_nmhandle_t *handle, isc_result_t eresult, void *cbarg) {
isc_nmhandle_t *readhandle = NULL;
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
UNUSED(cbarg);
F();
isc_refcount_decrement(&active_cconnects);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
if (eresult != ISC_R_SUCCESS) {
return;
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
}
atomic_fetch_add(&cconnects, 1);
isc_refcount_increment0(&active_creads);
isc_nmhandle_attach(handle, &readhandle);
isc_nm_read(handle, connect_read_cb, NULL);
connect_send(handle);
}
static void
listen_send_cb(isc_nmhandle_t *handle, isc_result_t eresult, void *cbarg) {
isc_nmhandle_t *sendhandle = handle;
UNUSED(cbarg);
UNUSED(eresult);
assert_non_null(sendhandle);
F();
if (eresult != ISC_R_SUCCESS) {
goto unref;
}
atomic_fetch_add(&ssends, 1);
unref:
isc_nmhandle_detach(&sendhandle);
isc_refcount_decrement(&active_ssends);
}
static void
listen_read_cb(isc_nmhandle_t *handle, isc_result_t eresult,
isc_region_t *region, void *cbarg) {
uint64_t magic = 0;
assert_non_null(handle);
F();
if (eresult != ISC_R_SUCCESS) {
goto unref;
}
atomic_fetch_add(&sreads, 1);
assert_int_equal(region->length, sizeof(magic));
memmove(&magic, region->base, sizeof(magic));
assert_true(magic == stop_magic || magic == send_magic);
if (magic == send_magic) {
isc_nmhandle_t *sendhandle = NULL;
isc_nmhandle_attach(handle, &sendhandle);
isc_refcount_increment0(&active_ssends);
isc_nm_send(sendhandle, (isc_region_t *)&send_msg,
listen_send_cb, cbarg);
return;
}
/* close the connection on stop_magic */
unref:
if (handle == cbarg) {
isc_refcount_decrement(&active_sreads);
isc_nmhandle_detach(&handle);
}
}
static isc_result_t
listen_accept_cb(isc_nmhandle_t *handle, isc_result_t eresult, void *cbarg) {
UNUSED(handle);
UNUSED(cbarg);
F();
return (eresult);
}
static isc_result_t
stream_accept_cb(isc_nmhandle_t *handle, isc_result_t eresult, void *cbarg) {
isc_nmhandle_t *readhandle = NULL;
UNUSED(cbarg);
F();
if (eresult != ISC_R_SUCCESS) {
return (eresult);
}
atomic_fetch_add(&saccepts, 1);
isc_refcount_increment0(&active_sreads);
isc_nmhandle_attach(handle, &readhandle);
isc_nm_read(handle, listen_read_cb, readhandle);
return (ISC_R_SUCCESS);
}
typedef void (*connect_func)(isc_nm_t *);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
static isc_threadresult_t
connect_thread(isc_threadarg_t arg) {
connect_func connect = (connect_func)arg;
isc_sockaddr_t connect_addr;
connect_addr = (isc_sockaddr_t){ .length = 0 };
isc_sockaddr_fromin6(&connect_addr, &in6addr_loopback, 0);
while (atomic_load(&do_send)) {
uint_fast32_t active =
isc_refcount_increment0(&active_cconnects);
if (active > workers) {
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
/*
* If we have more active connections than workers,
* start slowing down the connections to prevent the
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
* thundering herd problem.
*/
usleep((active - workers) * 1000);
}
connect(connect_nm);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
}
return ((isc_threadresult_t)0);
}
/* UDP */
static void
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
udp_connect(isc_nm_t *nm) {
isc_nm_udpconnect(nm, (isc_nmiface_t *)&udp_connect_addr,
(isc_nmiface_t *)&udp_listen_addr, connect_connect_cb,
NULL, T_CONNECT, 0);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
}
static void
mock_listenudp_uv_udp_open(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
WILL_RETURN(uv_udp_open, UV_ENOMEM);
result = isc_nm_listenudp(listen_nm, (isc_nmiface_t *)&udp_listen_addr,
noop_recv_cb, NULL, 0, &listen_sock);
assert_int_not_equal(result, ISC_R_SUCCESS);
assert_null(listen_sock);
RESET_RETURN;
}
static void
mock_listenudp_uv_udp_bind(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
WILL_RETURN(uv_udp_bind, UV_EADDRINUSE);
result = isc_nm_listenudp(listen_nm, (isc_nmiface_t *)&udp_listen_addr,
noop_recv_cb, NULL, 0, &listen_sock);
assert_int_not_equal(result, ISC_R_SUCCESS);
assert_null(listen_sock);
RESET_RETURN;
}
static void
mock_listenudp_uv_udp_recv_start(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
WILL_RETURN(uv_udp_recv_start, UV_EADDRINUSE);
result = isc_nm_listenudp(listen_nm, (isc_nmiface_t *)&udp_listen_addr,
noop_recv_cb, NULL, 0, &listen_sock);
assert_int_not_equal(result, ISC_R_SUCCESS);
assert_null(listen_sock);
RESET_RETURN;
}
static void
mock_udpconnect_uv_udp_open(void **state __attribute__((unused))) {
WILL_RETURN(uv_udp_open, UV_ENOMEM);
isc_refcount_increment0(&active_cconnects);
isc_nm_udpconnect(connect_nm, (isc_nmiface_t *)&udp_connect_addr,
(isc_nmiface_t *)&udp_listen_addr, noop_connect_cb,
NULL, T_CONNECT, 0);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
isc_nm_closedown(connect_nm);
RESET_RETURN;
}
static void
mock_udpconnect_uv_udp_bind(void **state __attribute__((unused))) {
WILL_RETURN(uv_udp_bind, UV_ENOMEM);
isc_refcount_increment0(&active_cconnects);
isc_nm_udpconnect(connect_nm, (isc_nmiface_t *)&udp_connect_addr,
(isc_nmiface_t *)&udp_listen_addr, noop_connect_cb,
NULL, T_CONNECT, 0);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
isc_nm_closedown(connect_nm);
RESET_RETURN;
}
#if HAVE_UV_UDP_CONNECT
static void
mock_udpconnect_uv_udp_connect(void **state __attribute__((unused))) {
WILL_RETURN(uv_udp_connect, UV_ENOMEM);
isc_refcount_increment0(&active_cconnects);
isc_nm_udpconnect(connect_nm, (isc_nmiface_t *)&udp_connect_addr,
(isc_nmiface_t *)&udp_listen_addr, noop_connect_cb,
NULL, T_CONNECT, 0);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
isc_nm_closedown(connect_nm);
RESET_RETURN;
}
#endif
static void
mock_udpconnect_uv_recv_buffer_size(void **state __attribute__((unused))) {
WILL_RETURN(uv_recv_buffer_size, UV_ENOMEM);
isc_refcount_increment0(&active_cconnects);
isc_nm_udpconnect(connect_nm, (isc_nmiface_t *)&udp_connect_addr,
(isc_nmiface_t *)&udp_listen_addr, noop_connect_cb,
NULL, T_CONNECT, 0);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
isc_nm_closedown(connect_nm);
RESET_RETURN;
}
static void
mock_udpconnect_uv_send_buffer_size(void **state __attribute__((unused))) {
WILL_RETURN(uv_send_buffer_size, UV_ENOMEM);
isc_refcount_increment0(&active_cconnects);
isc_nm_udpconnect(connect_nm, (isc_nmiface_t *)&udp_connect_addr,
(isc_nmiface_t *)&udp_listen_addr, noop_connect_cb,
NULL, T_CONNECT, 0);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
isc_nm_closedown(connect_nm);
RESET_RETURN;
}
static void
udp_noop(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
result = isc_nm_listenudp(listen_nm, (isc_nmiface_t *)&udp_listen_addr,
noop_recv_cb, NULL, 0, &listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
isc_refcount_increment0(&active_cconnects);
isc_nm_udpconnect(connect_nm, (isc_nmiface_t *)&udp_connect_addr,
(isc_nmiface_t *)&udp_listen_addr, noop_connect_cb,
NULL, T_CONNECT, 0);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
isc_nm_closedown(connect_nm);
atomic_assert_int_eq(cconnects, 0);
atomic_assert_int_eq(csends, 0);
atomic_assert_int_eq(creads, 0);
atomic_assert_int_eq(sreads, 0);
atomic_assert_int_eq(ssends, 0);
}
static void
udp_noresponse(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
result = isc_nm_listenudp(listen_nm, (isc_nmiface_t *)&udp_listen_addr,
noop_recv_cb, NULL, 0, &listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
isc_refcount_increment0(&active_cconnects);
isc_nm_udpconnect(connect_nm, (isc_nmiface_t *)&udp_connect_addr,
(isc_nmiface_t *)&udp_listen_addr, connect_connect_cb,
NULL, T_CONNECT, 0);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
WAIT_FOR_EQ(cconnects, 1);
WAIT_FOR_EQ(csends, 1);
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
isc_nm_closedown(connect_nm);
X(cconnects);
X(csends);
X(creads);
X(sreads);
X(ssends);
atomic_assert_int_eq(cconnects, 1);
atomic_assert_int_eq(csends, 1);
atomic_assert_int_eq(creads, 0);
atomic_assert_int_eq(sreads, 0);
atomic_assert_int_eq(ssends, 0);
}
static void
udp_recv_one(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
atomic_store(&nsends, 1);
result = isc_nm_listenudp(listen_nm, (isc_nmiface_t *)&udp_listen_addr,
listen_read_cb, NULL, 0, &listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
isc_refcount_increment0(&active_cconnects);
isc_nm_udpconnect(connect_nm, (isc_nmiface_t *)&udp_connect_addr,
(isc_nmiface_t *)&udp_listen_addr, connect_connect_cb,
NULL, T_CONNECT, 0);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
WAIT_FOR_EQ(cconnects, 1);
WAIT_FOR_LE(nsends, 0);
WAIT_FOR_EQ(csends, 1);
WAIT_FOR_EQ(sreads, 1);
WAIT_FOR_EQ(ssends, 0);
WAIT_FOR_EQ(creads, 0);
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
isc_nm_closedown(connect_nm);
X(cconnects);
X(csends);
X(creads);
X(sreads);
X(ssends);
atomic_assert_int_eq(cconnects, 1);
atomic_assert_int_eq(csends, 1);
atomic_assert_int_eq(creads, 0);
atomic_assert_int_eq(sreads, 1);
atomic_assert_int_eq(ssends, 0);
}
static void
udp_recv_two(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
atomic_store(&nsends, 2);
result = isc_nm_listenudp(listen_nm, (isc_nmiface_t *)&udp_listen_addr,
listen_read_cb, NULL, 0, &listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
isc_refcount_increment0(&active_cconnects);
isc_nm_udpconnect(connect_nm, (isc_nmiface_t *)&udp_connect_addr,
(isc_nmiface_t *)&udp_listen_addr, connect_connect_cb,
NULL, T_CONNECT, 0);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
WAIT_FOR_EQ(cconnects, 1);
isc_refcount_increment0(&active_cconnects);
isc_nm_udpconnect(connect_nm, (isc_nmiface_t *)&udp_connect_addr,
(isc_nmiface_t *)&udp_listen_addr, connect_connect_cb,
NULL, T_CONNECT, 0);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
WAIT_FOR_EQ(cconnects, 2);
WAIT_FOR_LE(nsends, 0);
WAIT_FOR_EQ(csends, 2);
WAIT_FOR_EQ(sreads, 2);
WAIT_FOR_EQ(ssends, 1);
WAIT_FOR_EQ(creads, 1);
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
isc_nm_closedown(connect_nm);
X(cconnects);
X(csends);
X(creads);
X(sreads);
X(ssends);
atomic_assert_int_eq(cconnects, 2);
atomic_assert_int_eq(csends, 2);
atomic_assert_int_eq(creads, 1);
atomic_assert_int_eq(sreads, 2);
atomic_assert_int_eq(ssends, 1);
}
static void
udp_recv_send(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
isc_thread_t threads[workers];
SKIP_IN_CI;
result = isc_nm_listenudp(listen_nm, (isc_nmiface_t *)&udp_listen_addr,
listen_read_cb, NULL, 0, &listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
memset(threads, 0, sizeof(threads));
for (size_t i = 0; i < workers; i++) {
isc_thread_create(connect_thread, udp_connect, &threads[i]);
}
WAIT_FOR_GE(cconnects, esends);
WAIT_FOR_GE(csends, esends);
WAIT_FOR_GE(sreads, esends);
WAIT_FOR_GE(ssends, esends / 2);
WAIT_FOR_GE(creads, esends / 2);
DONE();
for (size_t i = 0; i < workers; i++) {
isc_thread_join(threads[i], NULL);
}
isc_nm_closedown(connect_nm);
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
X(cconnects);
X(csends);
X(creads);
X(sreads);
X(ssends);
CHECK_RANGE_FULL(csends);
CHECK_RANGE_FULL(creads);
CHECK_RANGE_FULL(sreads);
CHECK_RANGE_FULL(ssends);
}
static void
udp_recv_half_send(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
isc_thread_t threads[workers];
SKIP_IN_CI;
result = isc_nm_listenudp(listen_nm, (isc_nmiface_t *)&udp_listen_addr,
listen_read_cb, NULL, 0, &listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
memset(threads, 0, sizeof(threads));
for (size_t i = 0; i < workers; i++) {
isc_thread_create(connect_thread, udp_connect, &threads[i]);
}
WAIT_FOR_GE(cconnects, esends / 2);
WAIT_FOR_GE(csends, esends / 2);
WAIT_FOR_GE(sreads, esends / 2);
WAIT_FOR_GE(ssends, esends / 2);
WAIT_FOR_GE(creads, esends / 2);
isc_nm_closedown(connect_nm);
DONE();
for (size_t i = 0; i < workers; i++) {
isc_thread_join(threads[i], NULL);
}
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
X(cconnects);
X(csends);
X(creads);
X(sreads);
X(ssends);
CHECK_RANGE_FULL(csends);
CHECK_RANGE_HALF(creads);
CHECK_RANGE_HALF(sreads);
CHECK_RANGE_HALF(ssends);
}
static void
udp_half_recv_send(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
isc_thread_t threads[workers];
SKIP_IN_CI;
result = isc_nm_listenudp(listen_nm, (isc_nmiface_t *)&udp_listen_addr,
listen_read_cb, NULL, 0, &listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
memset(threads, 0, sizeof(threads));
for (size_t i = 0; i < workers; i++) {
isc_thread_create(connect_thread, udp_connect, &threads[i]);
}
WAIT_FOR_GE(cconnects, esends / 2);
WAIT_FOR_GE(csends, esends / 2);
WAIT_FOR_GE(sreads, esends / 2);
WAIT_FOR_GE(ssends, esends / 2);
WAIT_FOR_GE(creads, esends / 2);
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
/* Try to send a little while longer */
usleep((esends / 2) * 10000);
isc_nm_closedown(connect_nm);
DONE();
for (size_t i = 0; i < workers; i++) {
isc_thread_join(threads[i], NULL);
}
X(cconnects);
X(csends);
X(creads);
X(sreads);
X(ssends);
CHECK_RANGE_FULL(csends);
CHECK_RANGE_HALF(creads);
CHECK_RANGE_HALF(sreads);
CHECK_RANGE_HALF(ssends);
}
static void
udp_half_recv_half_send(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
isc_thread_t threads[workers];
SKIP_IN_CI;
result = isc_nm_listenudp(listen_nm, (isc_nmiface_t *)&udp_listen_addr,
listen_read_cb, NULL, 0, &listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
memset(threads, 0, sizeof(threads));
for (size_t i = 0; i < workers; i++) {
isc_thread_create(connect_thread, udp_connect, &threads[i]);
}
WAIT_FOR_GE(cconnects, esends / 2);
WAIT_FOR_GE(csends, esends / 2);
WAIT_FOR_GE(sreads, esends / 2);
WAIT_FOR_GE(ssends, esends / 2);
WAIT_FOR_GE(creads, esends / 2);
isc_nm_closedown(connect_nm);
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
DONE();
for (size_t i = 0; i < workers; i++) {
isc_thread_join(threads[i], NULL);
}
X(cconnects);
X(csends);
X(creads);
X(sreads);
X(ssends);
CHECK_RANGE_FULL(csends);
CHECK_RANGE_HALF(creads);
CHECK_RANGE_HALF(sreads);
CHECK_RANGE_HALF(ssends);
}
/* TCP */
static isc_quota_t *
tcp_listener_init_quota(size_t nthreads);
static void
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
tcp_connect(isc_nm_t *nm) {
isc_nm_tcpconnect(nm, (isc_nmiface_t *)&tcp_connect_addr,
(isc_nmiface_t *)&tcp_listen_addr, connect_connect_cb,
NULL, 1, 0);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
}
static void
tcp_noop(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
result = isc_nm_listentcp(listen_nm, (isc_nmiface_t *)&tcp_listen_addr,
noop_accept_cb, NULL, 0, 0, NULL,
&listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
isc_refcount_increment0(&active_cconnects);
isc_nm_tcpconnect(connect_nm, (isc_nmiface_t *)&tcp_connect_addr,
(isc_nmiface_t *)&tcp_listen_addr, noop_connect_cb,
NULL, 1, 0);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
isc_nm_closedown(connect_nm);
atomic_assert_int_eq(cconnects, 0);
atomic_assert_int_eq(csends, 0);
atomic_assert_int_eq(creads, 0);
atomic_assert_int_eq(sreads, 0);
atomic_assert_int_eq(ssends, 0);
}
static void
tcp_noresponse(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
result = isc_nm_listentcp(listen_nm, (isc_nmiface_t *)&tcp_listen_addr,
noop_accept_cb, NULL, 0, 0, NULL,
&listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
isc_refcount_increment0(&active_cconnects);
isc_nm_tcpconnect(connect_nm, (isc_nmiface_t *)&tcp_connect_addr,
(isc_nmiface_t *)&tcp_listen_addr, connect_connect_cb,
NULL, 1, 0);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
WAIT_FOR_EQ(cconnects, 1);
WAIT_FOR_EQ(csends, 1);
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
isc_nm_closedown(connect_nm);
X(cconnects);
X(csends);
X(creads);
X(sreads);
X(ssends);
atomic_assert_int_eq(cconnects, 1);
atomic_assert_int_eq(csends, 1);
atomic_assert_int_eq(creads, 0);
atomic_assert_int_eq(sreads, 0);
atomic_assert_int_eq(ssends, 0);
}
static void
tcp_recv_one(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
isc_quota_t *quotap = tcp_listener_init_quota(1);
atomic_store(&nsends, 1);
result = isc_nm_listentcp(listen_nm, (isc_nmiface_t *)&tcp_listen_addr,
stream_accept_cb, NULL, 0, 0, quotap,
&listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
isc_refcount_increment0(&active_cconnects);
isc_nm_tcpconnect(connect_nm, (isc_nmiface_t *)&tcp_connect_addr,
(isc_nmiface_t *)&tcp_listen_addr, connect_connect_cb,
NULL, T_CONNECT, 0);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
WAIT_FOR_EQ(cconnects, 1);
WAIT_FOR_LE(nsends, 0);
WAIT_FOR_EQ(csends, 1);
WAIT_FOR_EQ(sreads, 1);
WAIT_FOR_EQ(ssends, 0);
WAIT_FOR_EQ(creads, 0);
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
isc_nm_closedown(connect_nm);
X(cconnects);
X(csends);
X(creads);
X(sreads);
X(ssends);
atomic_assert_int_eq(cconnects, 1);
atomic_assert_int_eq(csends, 1);
atomic_assert_int_eq(creads, 0);
atomic_assert_int_eq(sreads, 1);
atomic_assert_int_eq(ssends, 0);
}
static void
tcp_recv_two(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
isc_quota_t *quotap = tcp_listener_init_quota(1);
atomic_store(&nsends, 2);
result = isc_nm_listentcp(listen_nm, (isc_nmiface_t *)&tcp_listen_addr,
stream_accept_cb, NULL, 0, 0, quotap,
&listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
isc_refcount_increment0(&active_cconnects);
isc_nm_tcpconnect(connect_nm, (isc_nmiface_t *)&tcp_connect_addr,
(isc_nmiface_t *)&tcp_listen_addr, connect_connect_cb,
NULL, T_CONNECT, 0);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
WAIT_FOR_EQ(cconnects, 1);
isc_refcount_increment0(&active_cconnects);
isc_nm_tcpconnect(connect_nm, (isc_nmiface_t *)&tcp_connect_addr,
(isc_nmiface_t *)&tcp_listen_addr, connect_connect_cb,
NULL, T_CONNECT, 0);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
WAIT_FOR_EQ(cconnects, 2);
WAIT_FOR_LE(nsends, 0);
WAIT_FOR_EQ(csends, 2);
WAIT_FOR_EQ(sreads, 2);
WAIT_FOR_EQ(ssends, 1);
WAIT_FOR_EQ(creads, 1);
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
isc_nm_closedown(connect_nm);
X(cconnects);
X(csends);
X(creads);
X(sreads);
X(ssends);
atomic_assert_int_eq(cconnects, 2);
atomic_assert_int_eq(csends, 2);
atomic_assert_int_eq(creads, 1);
atomic_assert_int_eq(sreads, 2);
atomic_assert_int_eq(ssends, 1);
}
static void
tcp_recv_send(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
isc_thread_t threads[workers];
isc_quota_t *quotap = tcp_listener_init_quota(workers);
SKIP_IN_CI;
result = isc_nm_listentcp(listen_nm, (isc_nmiface_t *)&tcp_listen_addr,
stream_accept_cb, NULL, 0, 0, quotap,
&listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
memset(threads, 0, sizeof(threads));
for (size_t i = 0; i < workers; i++) {
isc_thread_create(connect_thread, tcp_connect, &threads[i]);
}
WAIT_FOR_GE(cconnects, esends);
WAIT_FOR_GE(csends, esends);
WAIT_FOR_GE(sreads, esends);
WAIT_FOR_GE(ssends, esends / 2);
WAIT_FOR_GE(creads, esends / 2);
DONE();
for (size_t i = 0; i < workers; i++) {
isc_thread_join(threads[i], NULL);
}
isc_nm_closedown(connect_nm);
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
X(cconnects);
X(csends);
X(creads);
X(sreads);
X(ssends);
CHECK_RANGE_FULL(csends);
CHECK_RANGE_FULL(creads);
CHECK_RANGE_FULL(sreads);
CHECK_RANGE_FULL(ssends);
}
static void
tcp_recv_half_send(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
isc_thread_t threads[workers];
isc_quota_t *quotap = tcp_listener_init_quota(workers);
SKIP_IN_CI;
result = isc_nm_listentcp(listen_nm, (isc_nmiface_t *)&tcp_listen_addr,
stream_accept_cb, NULL, 0, 0, quotap,
&listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
memset(threads, 0, sizeof(threads));
for (size_t i = 0; i < workers; i++) {
isc_thread_create(connect_thread, tcp_connect, &threads[i]);
}
WAIT_FOR_GE(cconnects, esends / 2);
WAIT_FOR_GE(csends, esends / 2);
WAIT_FOR_GE(sreads, esends / 2);
WAIT_FOR_GE(ssends, esends / 2);
WAIT_FOR_GE(creads, esends / 2);
isc_nm_closedown(connect_nm);
DONE();
for (size_t i = 0; i < workers; i++) {
isc_thread_join(threads[i], NULL);
}
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
X(cconnects);
X(csends);
X(creads);
X(sreads);
X(ssends);
CHECK_RANGE_HALF(csends);
CHECK_RANGE_HALF(creads);
CHECK_RANGE_HALF(sreads);
CHECK_RANGE_HALF(ssends);
}
static void
tcp_half_recv_send(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
isc_thread_t threads[workers];
isc_quota_t *quotap = tcp_listener_init_quota(workers);
SKIP_IN_CI;
result = isc_nm_listentcp(listen_nm, (isc_nmiface_t *)&tcp_listen_addr,
stream_accept_cb, NULL, 0, 0, quotap,
&listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
memset(threads, 0, sizeof(threads));
for (size_t i = 0; i < workers; i++) {
isc_thread_create(connect_thread, tcp_connect, &threads[i]);
}
WAIT_FOR_GE(cconnects, esends / 2);
WAIT_FOR_GE(csends, esends / 2);
WAIT_FOR_GE(sreads, esends / 2);
WAIT_FOR_GE(ssends, esends / 2);
WAIT_FOR_GE(creads, esends / 2);
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
/* Try to send a little while longer */
usleep((esends / 2) * 10000);
isc_nm_closedown(connect_nm);
DONE();
for (size_t i = 0; i < workers; i++) {
isc_thread_join(threads[i], NULL);
}
X(cconnects);
X(csends);
X(creads);
X(sreads);
X(ssends);
CHECK_RANGE_HALF(csends);
CHECK_RANGE_HALF(creads);
CHECK_RANGE_HALF(sreads);
CHECK_RANGE_HALF(ssends);
}
static void
tcp_half_recv_half_send(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
isc_thread_t threads[workers];
isc_quota_t *quotap = tcp_listener_init_quota(workers);
SKIP_IN_CI;
result = isc_nm_listentcp(listen_nm, (isc_nmiface_t *)&tcp_listen_addr,
stream_accept_cb, NULL, 0, 0, quotap,
&listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
memset(threads, 0, sizeof(threads));
for (size_t i = 0; i < workers; i++) {
isc_thread_create(connect_thread, tcp_connect, &threads[i]);
}
WAIT_FOR_GE(cconnects, esends / 2);
WAIT_FOR_GE(csends, esends / 2);
WAIT_FOR_GE(sreads, esends / 2);
WAIT_FOR_GE(ssends, esends / 2);
WAIT_FOR_GE(creads, esends / 2);
isc_nm_closedown(connect_nm);
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
DONE();
for (size_t i = 0; i < workers; i++) {
isc_thread_join(threads[i], NULL);
}
X(cconnects);
X(csends);
X(creads);
X(sreads);
X(ssends);
CHECK_RANGE_HALF(csends);
CHECK_RANGE_HALF(creads);
CHECK_RANGE_HALF(sreads);
CHECK_RANGE_HALF(ssends);
}
/* TCP Quota */
static isc_quota_t *
tcp_listener_init_quota(size_t nthreads) {
isc_quota_t *quotap = NULL;
if (atomic_load(&check_listener_quota)) {
unsigned max_quota = ISC_MAX(nthreads / 2, 1);
isc_quota_max(&listener_quota, max_quota);
quotap = &listener_quota;
}
return (quotap);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
}
static void
tcp_recv_one_quota(void **state) {
atomic_store(&check_listener_quota, true);
tcp_recv_one(state);
}
static void
tcp_recv_two_quota(void **state) {
atomic_store(&check_listener_quota, true);
tcp_recv_two(state);
}
static void
tcp_recv_send_quota(void **state) {
SKIP_IN_CI;
atomic_store(&check_listener_quota, true);
tcp_recv_send(state);
}
static void
tcp_recv_half_send_quota(void **state) {
SKIP_IN_CI;
atomic_store(&check_listener_quota, true);
tcp_recv_half_send(state);
}
static void
tcp_half_recv_send_quota(void **state) {
SKIP_IN_CI;
atomic_store(&check_listener_quota, true);
tcp_half_recv_send(state);
}
static void
tcp_half_recv_half_send_quota(void **state) {
SKIP_IN_CI;
atomic_store(&check_listener_quota, true);
tcp_half_recv_half_send(state);
}
/* TCPDNS */
static void
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
tcpdns_connect(isc_nm_t *nm) {
isc_nm_tcpdnsconnect(nm, (isc_nmiface_t *)&tcp_connect_addr,
(isc_nmiface_t *)&tcp_listen_addr,
connect_connect_cb, NULL, T_CONNECT, 0);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
}
static void
tcpdns_noop(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
result = isc_nm_listentcpdns(
listen_nm, (isc_nmiface_t *)&tcp_listen_addr, noop_recv_cb,
NULL, noop_accept_cb, NULL, 0, 0, NULL, &listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
isc_refcount_increment0(&active_cconnects);
isc_nm_tcpdnsconnect(connect_nm, (isc_nmiface_t *)&tcp_connect_addr,
(isc_nmiface_t *)&tcp_listen_addr, noop_connect_cb,
NULL, T_CONNECT, 0);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
isc_nm_closedown(connect_nm);
atomic_assert_int_eq(cconnects, 0);
atomic_assert_int_eq(csends, 0);
atomic_assert_int_eq(creads, 0);
atomic_assert_int_eq(sreads, 0);
atomic_assert_int_eq(ssends, 0);
}
static void
tcpdns_noresponse(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
isc_refcount_increment0(&active_cconnects);
result = isc_nm_listentcpdns(
listen_nm, (isc_nmiface_t *)&tcp_listen_addr, noop_recv_cb,
NULL, noop_accept_cb, NULL, 0, 0, NULL, &listen_sock);
if (result != ISC_R_SUCCESS) {
isc_refcount_decrement(&active_cconnects);
usleep(1000);
}
assert_int_equal(result, ISC_R_SUCCESS);
isc_nm_tcpdnsconnect(connect_nm, (isc_nmiface_t *)&tcp_connect_addr,
(isc_nmiface_t *)&tcp_listen_addr,
connect_connect_cb, NULL, T_CONNECT, 0);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
WAIT_FOR_EQ(cconnects, 1);
WAIT_FOR_EQ(csends, 1);
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
isc_nm_closedown(connect_nm);
X(cconnects);
X(csends);
X(creads);
X(sreads);
X(ssends);
atomic_assert_int_eq(cconnects, 1);
atomic_assert_int_eq(csends, 1);
atomic_assert_int_eq(creads, 0);
atomic_assert_int_eq(sreads, 0);
atomic_assert_int_eq(ssends, 0);
}
static void
tcpdns_recv_one(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
atomic_store(&nsends, 1);
result = isc_nm_listentcpdns(
listen_nm, (isc_nmiface_t *)&tcp_listen_addr, listen_read_cb,
NULL, listen_accept_cb, NULL, 0, 0, NULL, &listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
isc_refcount_increment0(&active_cconnects);
isc_nm_tcpdnsconnect(connect_nm, (isc_nmiface_t *)&tcp_connect_addr,
(isc_nmiface_t *)&tcp_listen_addr,
connect_connect_cb, NULL, T_CONNECT, 0);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
WAIT_FOR_EQ(cconnects, 1);
WAIT_FOR_LE(nsends, 0);
WAIT_FOR_EQ(csends, 1);
WAIT_FOR_EQ(sreads, 1);
WAIT_FOR_EQ(ssends, 0);
WAIT_FOR_EQ(creads, 0);
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
isc_nm_closedown(connect_nm);
X(cconnects);
X(csends);
X(creads);
X(sreads);
X(ssends);
atomic_assert_int_eq(cconnects, 1);
atomic_assert_int_eq(csends, 1);
atomic_assert_int_eq(creads, 0);
atomic_assert_int_eq(sreads, 1);
atomic_assert_int_eq(ssends, 0);
}
static void
tcpdns_recv_two(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
atomic_store(&nsends, 2);
result = isc_nm_listentcpdns(
listen_nm, (isc_nmiface_t *)&tcp_listen_addr, listen_read_cb,
NULL, listen_accept_cb, NULL, 0, 0, NULL, &listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
isc_refcount_increment0(&active_cconnects);
isc_nm_tcpdnsconnect(connect_nm, (isc_nmiface_t *)&tcp_connect_addr,
(isc_nmiface_t *)&tcp_listen_addr,
connect_connect_cb, NULL, T_CONNECT, 0);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
WAIT_FOR_EQ(cconnects, 1);
isc_refcount_increment0(&active_cconnects);
isc_nm_tcpdnsconnect(connect_nm, (isc_nmiface_t *)&tcp_connect_addr,
(isc_nmiface_t *)&tcp_listen_addr,
connect_connect_cb, NULL, T_CONNECT, 0);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
WAIT_FOR_EQ(cconnects, 2);
WAIT_FOR_LE(nsends, 0);
WAIT_FOR_EQ(csends, 2);
WAIT_FOR_EQ(sreads, 2);
WAIT_FOR_EQ(ssends, 1);
WAIT_FOR_EQ(creads, 1);
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
isc_nm_closedown(connect_nm);
X(cconnects);
X(csends);
X(creads);
X(sreads);
X(ssends);
atomic_assert_int_eq(cconnects, 2);
atomic_assert_int_eq(csends, 2);
atomic_assert_int_eq(creads, 1);
atomic_assert_int_eq(sreads, 2);
atomic_assert_int_eq(ssends, 1);
}
static void
tcpdns_recv_send(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
isc_thread_t threads[workers];
SKIP_IN_CI;
result = isc_nm_listentcpdns(
listen_nm, (isc_nmiface_t *)&tcp_listen_addr, listen_read_cb,
NULL, listen_accept_cb, NULL, 0, 0, NULL, &listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
memset(threads, 0, sizeof(threads));
for (size_t i = 0; i < workers; i++) {
isc_thread_create(connect_thread, tcpdns_connect, &threads[i]);
}
WAIT_FOR_GE(cconnects, esends);
WAIT_FOR_GE(csends, esends);
WAIT_FOR_GE(sreads, esends);
WAIT_FOR_GE(ssends, esends / 2);
WAIT_FOR_GE(creads, esends / 2);
DONE();
for (size_t i = 0; i < workers; i++) {
isc_thread_join(threads[i], NULL);
}
isc_nm_closedown(connect_nm);
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
X(cconnects);
X(csends);
X(creads);
X(sreads);
X(ssends);
CHECK_RANGE_FULL(csends);
CHECK_RANGE_FULL(creads);
CHECK_RANGE_FULL(sreads);
CHECK_RANGE_FULL(ssends);
}
static void
tcpdns_recv_half_send(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
isc_thread_t threads[workers];
SKIP_IN_CI;
result = isc_nm_listentcpdns(
listen_nm, (isc_nmiface_t *)&tcp_listen_addr, listen_read_cb,
NULL, listen_accept_cb, NULL, 0, 0, NULL, &listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
memset(threads, 0, sizeof(threads));
for (size_t i = 0; i < workers; i++) {
isc_thread_create(connect_thread, tcpdns_connect, &threads[i]);
}
WAIT_FOR_GE(cconnects, esends / 2);
WAIT_FOR_GE(csends, esends / 2);
WAIT_FOR_GE(sreads, esends / 2);
WAIT_FOR_GE(ssends, esends / 2);
WAIT_FOR_GE(creads, esends / 2);
isc_nm_closedown(connect_nm);
DONE();
for (size_t i = 0; i < workers; i++) {
isc_thread_join(threads[i], NULL);
}
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
X(cconnects);
X(csends);
X(creads);
X(sreads);
X(ssends);
CHECK_RANGE_HALF(csends);
CHECK_RANGE_HALF(creads);
CHECK_RANGE_HALF(sreads);
CHECK_RANGE_HALF(ssends);
}
static void
tcpdns_half_recv_send(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
isc_thread_t threads[workers];
SKIP_IN_CI;
result = isc_nm_listentcpdns(
listen_nm, (isc_nmiface_t *)&tcp_listen_addr, listen_read_cb,
NULL, listen_accept_cb, NULL, 0, 0, NULL, &listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
memset(threads, 0, sizeof(threads));
for (size_t i = 0; i < workers; i++) {
isc_thread_create(connect_thread, tcpdns_connect, &threads[i]);
}
WAIT_FOR_GE(cconnects, esends / 2);
WAIT_FOR_GE(csends, esends / 2);
WAIT_FOR_GE(sreads, esends / 2);
WAIT_FOR_GE(ssends, esends / 2);
WAIT_FOR_GE(creads, esends / 2);
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
/* Try to send a little while longer */
usleep((esends / 2) * 10000);
isc_nm_closedown(connect_nm);
DONE();
for (size_t i = 0; i < workers; i++) {
isc_thread_join(threads[i], NULL);
}
X(cconnects);
X(csends);
X(creads);
X(sreads);
X(ssends);
CHECK_RANGE_HALF(csends);
CHECK_RANGE_HALF(creads);
CHECK_RANGE_HALF(sreads);
CHECK_RANGE_HALF(ssends);
}
static void
tcpdns_half_recv_half_send(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
isc_thread_t threads[workers];
SKIP_IN_CI;
result = isc_nm_listentcpdns(
listen_nm, (isc_nmiface_t *)&tcp_listen_addr, listen_read_cb,
NULL, listen_accept_cb, NULL, 0, 0, NULL, &listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
memset(threads, 0, sizeof(threads));
for (size_t i = 0; i < workers; i++) {
isc_thread_create(connect_thread, tcpdns_connect, &threads[i]);
}
WAIT_FOR_GE(cconnects, esends / 2);
WAIT_FOR_GE(csends, esends / 2);
WAIT_FOR_GE(sreads, esends / 2);
WAIT_FOR_GE(ssends, esends / 2);
WAIT_FOR_GE(creads, esends / 2);
isc_nm_closedown(connect_nm);
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
DONE();
for (size_t i = 0; i < workers; i++) {
isc_thread_join(threads[i], NULL);
}
X(cconnects);
X(csends);
X(creads);
X(sreads);
X(ssends);
CHECK_RANGE_HALF(csends);
CHECK_RANGE_HALF(creads);
CHECK_RANGE_HALF(sreads);
CHECK_RANGE_HALF(ssends);
}
/* TLSDNS */
static void
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
tlsdns_connect(isc_nm_t *nm) {
isc_nm_tlsdnsconnect(nm, (isc_nmiface_t *)&tcp_connect_addr,
(isc_nmiface_t *)&tcp_listen_addr,
connect_connect_cb, NULL, T_CONNECT, 0,
tcp_connect_tlsctx);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
}
static void
tlsdns_noop(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
result = isc_nm_listentlsdns(
listen_nm, (isc_nmiface_t *)&tcp_listen_addr, noop_recv_cb,
NULL, noop_accept_cb, NULL, 0, 0, NULL, tcp_listen_tlsctx,
&listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
isc_refcount_increment0(&active_cconnects);
isc_nm_tlsdnsconnect(connect_nm, (isc_nmiface_t *)&tcp_connect_addr,
(isc_nmiface_t *)&tcp_listen_addr, noop_connect_cb,
NULL, T_CONNECT, 0, tcp_connect_tlsctx);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
isc_nm_closedown(connect_nm);
atomic_assert_int_eq(cconnects, 0);
atomic_assert_int_eq(csends, 0);
atomic_assert_int_eq(creads, 0);
atomic_assert_int_eq(sreads, 0);
atomic_assert_int_eq(ssends, 0);
}
static void
tlsdns_noresponse(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
isc_sockaddr_t connect_addr;
connect_addr = (isc_sockaddr_t){ .length = 0 };
isc_sockaddr_fromin6(&connect_addr, &in6addr_loopback, 0);
result = isc_nm_listentlsdns(
listen_nm, (isc_nmiface_t *)&tcp_listen_addr, noop_recv_cb,
NULL, noop_accept_cb, NULL, 0, 0, NULL, tcp_listen_tlsctx,
&listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
isc_refcount_increment0(&active_cconnects);
isc_nm_tlsdnsconnect(connect_nm, (isc_nmiface_t *)&connect_addr,
(isc_nmiface_t *)&tcp_listen_addr,
connect_connect_cb, NULL, T_CONNECT, 0,
tcp_connect_tlsctx);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
WAIT_FOR_EQ(cconnects, 1);
WAIT_FOR_EQ(csends, 1);
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
isc_nm_closedown(connect_nm);
X(cconnects);
X(csends);
X(creads);
X(sreads);
X(ssends);
atomic_assert_int_eq(cconnects, 1);
atomic_assert_int_eq(csends, 1);
atomic_assert_int_eq(creads, 0);
atomic_assert_int_eq(sreads, 0);
atomic_assert_int_eq(ssends, 0);
}
static void
tlsdns_recv_one(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
atomic_store(&nsends, 1);
result = isc_nm_listentlsdns(
listen_nm, (isc_nmiface_t *)&tcp_listen_addr, listen_read_cb,
NULL, listen_accept_cb, NULL, 0, 0, NULL, tcp_listen_tlsctx,
&listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
isc_refcount_increment0(&active_cconnects);
isc_nm_tlsdnsconnect(connect_nm, (isc_nmiface_t *)&tcp_connect_addr,
(isc_nmiface_t *)&tcp_listen_addr,
connect_connect_cb, NULL, T_CONNECT, 0,
tcp_connect_tlsctx);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
WAIT_FOR_EQ(cconnects, 1);
WAIT_FOR_LE(nsends, 0);
WAIT_FOR_EQ(csends, 1);
WAIT_FOR_EQ(sreads, 1);
WAIT_FOR_EQ(ssends, 0);
WAIT_FOR_EQ(creads, 0);
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
isc_nm_closedown(connect_nm);
X(cconnects);
X(csends);
X(creads);
X(sreads);
X(ssends);
atomic_assert_int_eq(cconnects, 1);
atomic_assert_int_eq(csends, 1);
atomic_assert_int_eq(creads, 0);
atomic_assert_int_eq(sreads, 1);
atomic_assert_int_eq(ssends, 0);
}
static void
tlsdns_recv_two(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
atomic_store(&nsends, 2);
result = isc_nm_listentlsdns(
listen_nm, (isc_nmiface_t *)&tcp_listen_addr, listen_read_cb,
NULL, listen_accept_cb, NULL, 0, 0, NULL, tcp_listen_tlsctx,
&listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
isc_refcount_increment0(&active_cconnects);
isc_nm_tlsdnsconnect(connect_nm, (isc_nmiface_t *)&tcp_connect_addr,
(isc_nmiface_t *)&tcp_listen_addr,
connect_connect_cb, NULL, T_CONNECT, 0,
tcp_connect_tlsctx);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
WAIT_FOR_EQ(cconnects, 1);
isc_refcount_increment0(&active_cconnects);
isc_nm_tlsdnsconnect(connect_nm, (isc_nmiface_t *)&tcp_connect_addr,
(isc_nmiface_t *)&tcp_listen_addr,
connect_connect_cb, NULL, T_CONNECT, 0,
tcp_connect_tlsctx);
Improve reliability of the netmgr unit tests The netmgr unit tests were designed to push the system limits to maximum by sending as many queries as possible in the busy loop from multiple threads. This mostly works with UDP, but in the stateful protocol where establishing the connection takes more time, it failed quite often in the CI. On FreeBSD, this happened more often, because the socket() call would fail spuriosly making the problem even worse. This commit does several things to improve reliability: * return value of isc_nm_<proto>connect() is always checked and retried when scheduling the connection fails * The busy while loop has been slowed down with usleep(1000); so the netmgr threads could schedule the work and get executed. * The isc_thread_yield() was replaced with usleep(1000); also to allow the other threads to do any work. * Instead of waiting on just one variable, we wait for multiple variables to reach the final value * We are wrapping the netmgr operations (connects, reads, writes, accepts) with reference counting and waiting for all the callbacks to be accounted for. This has two effects: a) the isc_nm_t is always clean of active sockets and handles when destroyed, so it will prevent the spurious INSIST(references == 1) from isc_nm_destroy() b) the unit test now ensures that all the callbacks are always called when they should be called, so any stuck test means that there was a missing callback call and it is always a real bug These changes allows us to remove the workaround that would not run certain tests on systems without port load-balancing.
2021-01-29 13:00:46 +01:00
WAIT_FOR_EQ(cconnects, 2);
WAIT_FOR_LE(nsends, 0);
WAIT_FOR_EQ(csends, 2);
WAIT_FOR_EQ(sreads, 2);
WAIT_FOR_EQ(ssends, 1);
WAIT_FOR_EQ(creads, 1);
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
isc_nm_closedown(connect_nm);
X(cconnects);
X(csends);
X(creads);
X(sreads);
X(ssends);
atomic_assert_int_eq(cconnects, 2);
atomic_assert_int_eq(csends, 2);
atomic_assert_int_eq(creads, 1);
atomic_assert_int_eq(sreads, 2);
atomic_assert_int_eq(ssends, 1);
}
static void
tlsdns_recv_send(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
isc_thread_t threads[workers];
SKIP_IN_CI;
result = isc_nm_listentlsdns(
listen_nm, (isc_nmiface_t *)&tcp_listen_addr, listen_read_cb,
NULL, listen_accept_cb, NULL, 0, 0, NULL, tcp_listen_tlsctx,
&listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
memset(threads, 0, sizeof(threads));
for (size_t i = 0; i < workers; i++) {
isc_thread_create(connect_thread, tlsdns_connect, &threads[i]);
}
WAIT_FOR_GE(cconnects, esends);
WAIT_FOR_GE(csends, esends);
WAIT_FOR_GE(sreads, esends);
WAIT_FOR_GE(ssends, esends / 2);
WAIT_FOR_GE(creads, esends / 2);
DONE();
for (size_t i = 0; i < workers; i++) {
isc_thread_join(threads[i], NULL);
}
isc_nm_closedown(connect_nm);
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
X(cconnects);
X(csends);
X(creads);
X(sreads);
X(ssends);
CHECK_RANGE_FULL(csends);
CHECK_RANGE_FULL(creads);
CHECK_RANGE_FULL(sreads);
CHECK_RANGE_FULL(ssends);
}
static void
tlsdns_recv_half_send(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
isc_thread_t threads[workers];
SKIP_IN_CI;
result = isc_nm_listentlsdns(
listen_nm, (isc_nmiface_t *)&tcp_listen_addr, listen_read_cb,
NULL, listen_accept_cb, NULL, 0, 0, NULL, tcp_listen_tlsctx,
&listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
memset(threads, 0, sizeof(threads));
for (size_t i = 0; i < workers; i++) {
isc_thread_create(connect_thread, tlsdns_connect, &threads[i]);
}
WAIT_FOR_GE(cconnects, esends / 2);
WAIT_FOR_GE(csends, esends / 2);
WAIT_FOR_GE(sreads, esends / 2);
WAIT_FOR_GE(ssends, esends / 2);
WAIT_FOR_GE(creads, esends / 2);
isc_nm_closedown(connect_nm);
DONE();
for (size_t i = 0; i < workers; i++) {
isc_thread_join(threads[i], NULL);
}
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
X(cconnects);
X(csends);
X(creads);
X(sreads);
X(ssends);
CHECK_RANGE_HALF(csends);
CHECK_RANGE_HALF(creads);
CHECK_RANGE_HALF(sreads);
CHECK_RANGE_HALF(ssends);
}
static void
tlsdns_half_recv_send(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
isc_thread_t threads[workers];
SKIP_IN_CI;
result = isc_nm_listentlsdns(
listen_nm, (isc_nmiface_t *)&tcp_listen_addr, listen_read_cb,
NULL, listen_accept_cb, NULL, 0, 0, NULL, tcp_listen_tlsctx,
&listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
memset(threads, 0, sizeof(threads));
for (size_t i = 0; i < workers; i++) {
isc_thread_create(connect_thread, tlsdns_connect, &threads[i]);
}
WAIT_FOR_GE(cconnects, esends / 2);
WAIT_FOR_GE(csends, esends / 2);
WAIT_FOR_GE(sreads, esends / 2);
WAIT_FOR_GE(ssends, esends / 2);
WAIT_FOR_GE(creads, esends / 2);
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
/* Try to send a little while longer */
usleep((esends / 2) * 10000);
isc_nm_closedown(connect_nm);
DONE();
for (size_t i = 0; i < workers; i++) {
isc_thread_join(threads[i], NULL);
}
X(cconnects);
X(csends);
X(creads);
X(sreads);
X(ssends);
CHECK_RANGE_HALF(csends);
CHECK_RANGE_HALF(creads);
CHECK_RANGE_HALF(sreads);
CHECK_RANGE_HALF(ssends);
}
static void
tlsdns_half_recv_half_send(void **state __attribute__((unused))) {
isc_result_t result = ISC_R_SUCCESS;
isc_nmsocket_t *listen_sock = NULL;
isc_thread_t threads[workers];
SKIP_IN_CI;
result = isc_nm_listentlsdns(
listen_nm, (isc_nmiface_t *)&tcp_listen_addr, listen_read_cb,
NULL, listen_accept_cb, NULL, 0, 0, NULL, tcp_listen_tlsctx,
&listen_sock);
assert_int_equal(result, ISC_R_SUCCESS);
memset(threads, 0, sizeof(threads));
for (size_t i = 0; i < workers; i++) {
isc_thread_create(connect_thread, tlsdns_connect, &threads[i]);
}
WAIT_FOR_GE(cconnects, esends / 2);
WAIT_FOR_GE(csends, esends / 2);
WAIT_FOR_GE(sreads, esends / 2);
WAIT_FOR_GE(ssends, esends / 2);
WAIT_FOR_GE(creads, esends / 2);
isc_nm_closedown(connect_nm);
isc_nm_stoplistening(listen_sock);
isc_nmsocket_close(&listen_sock);
assert_null(listen_sock);
DONE();
for (size_t i = 0; i < workers; i++) {
isc_thread_join(threads[i], NULL);
}
X(cconnects);
X(csends);
X(creads);
X(sreads);
X(ssends);
CHECK_RANGE_HALF(csends);
CHECK_RANGE_HALF(creads);
CHECK_RANGE_HALF(sreads);
CHECK_RANGE_HALF(ssends);
}
int
main(void) {
const struct CMUnitTest tests[] = {
/* UDP */
cmocka_unit_test_setup_teardown(mock_listenudp_uv_udp_open,
nm_setup, nm_teardown),
cmocka_unit_test_setup_teardown(mock_listenudp_uv_udp_bind,
nm_setup, nm_teardown),
cmocka_unit_test_setup_teardown(
mock_listenudp_uv_udp_recv_start, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(mock_udpconnect_uv_udp_open,
nm_setup, nm_teardown),
cmocka_unit_test_setup_teardown(mock_udpconnect_uv_udp_bind,
nm_setup, nm_teardown),
#if HAVE_UV_UDP_CONNECT
cmocka_unit_test_setup_teardown(mock_udpconnect_uv_udp_connect,
nm_setup, nm_teardown),
#endif
cmocka_unit_test_setup_teardown(
mock_udpconnect_uv_recv_buffer_size, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(
mock_udpconnect_uv_send_buffer_size, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(udp_noop, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(udp_noresponse, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(udp_recv_one, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(udp_recv_two, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(udp_recv_send, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(udp_recv_half_send, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(udp_half_recv_send, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(udp_half_recv_half_send,
nm_setup, nm_teardown),
/* TCP */
/* cmocka_unit_test_setup_teardown(mock_listentcp_uv_tcp_bind,
*/
/* nm_setup, nm_teardown), */
/* cmocka_unit_test_setup_teardown(mock_listentcp_uv_fileno, */
/* nm_setup, nm_teardown), */
/* cmocka_unit_test_setup_teardown( */
/* mock_listentcp_uv_tcp_getsockname, nm_setup, */
/* nm_teardown), */
/* cmocka_unit_test_setup_teardown(mock_listentcp_uv_listen, */
/* nm_setup, nm_teardown), */
/* cmocka_unit_test_setup_teardown(mock_tcpconnect_uv_tcp_bind,
*/
/* nm_setup, nm_teardown), */
/* cmocka_unit_test_setup_teardown(mock_tcpconnect_uv_tcp_connect,
*/
/* nm_setup, nm_teardown), */
cmocka_unit_test_setup_teardown(tcp_noop, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(tcp_noresponse, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(tcp_recv_one, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(tcp_recv_two, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(tcp_recv_send, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(tcp_recv_half_send, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(tcp_half_recv_send, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(tcp_half_recv_half_send,
nm_setup, nm_teardown),
/* TCP Quota */
cmocka_unit_test_setup_teardown(tcp_recv_one_quota, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(tcp_recv_two_quota, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(tcp_recv_send_quota, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(tcp_recv_half_send_quota,
nm_setup, nm_teardown),
cmocka_unit_test_setup_teardown(tcp_half_recv_send_quota,
nm_setup, nm_teardown),
cmocka_unit_test_setup_teardown(tcp_half_recv_half_send_quota,
nm_setup, nm_teardown),
/* TCPDNS */
cmocka_unit_test_setup_teardown(tcpdns_recv_one, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(tcpdns_recv_two, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(tcpdns_noop, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(tcpdns_noresponse, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(tcpdns_recv_send, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(tcpdns_recv_half_send, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(tcpdns_half_recv_send, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(tcpdns_half_recv_half_send,
nm_setup, nm_teardown),
/* TLSDNS */
cmocka_unit_test_setup_teardown(tlsdns_recv_one, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(tlsdns_recv_two, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(tlsdns_noop, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(tlsdns_noresponse, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(tlsdns_recv_send, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(tlsdns_recv_half_send, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(tlsdns_half_recv_send, nm_setup,
nm_teardown),
cmocka_unit_test_setup_teardown(tlsdns_half_recv_half_send,
nm_setup, nm_teardown),
};
return (cmocka_run_group_tests(tests, _setup, _teardown));
}
#else /* HAVE_CMOCKA */
#include <stdio.h>
int
main(void) {
printf("1..0 # Skipped: cmocka not available\n");
return (SKIPPED_TEST_EXIT_CODE);
}
#endif /* if HAVE_CMOCKA */