myxce/src/lib/testsuite/test_netio_bounded_connect.c

/*
 * test_netio_bounded_connect.c — TDD tests for the bounded peer connect path.
 *
 * Task 1.3: raft-drive-async spec
 * Boundary: BoundedConnect (netio.c / netio.h)
 * Requirement: 2.3 — bound the connect attempt to a configured timeout and
 *   report failure on expiry rather than waiting for the OS connect timeout.
 *
 * Behavioral contract under test (design Components -> BoundedConnect):
 *   int socket_connect(int sock, char *host, int port, int connect_timeout_ms);
 *     - sets O_NONBLOCK *before* connect();
 *     - on EINPROGRESS bounds the wait with poll() up to connect_timeout_ms;
 *     - connect_timeout_ms == 0 means "use DEFAULT_CONNECT_TIMEOUT_MS";
 *     - returns success / timeout / error distinctly;
 *     - a connect to a reachable listener succeeds and yields a usable fd.
 *
 * Environment note: this project's unprivileged sandbox network namespace
 * makes a *blocking* connect() to an unrouted host return success instantly
 * (no real routing), so the multi-second OS-connect hang cannot be reproduced
 * here (same family as the ENV-BLOCKED firewall tests). These tests therefore
 * assert the deterministic, environment-portable properties of the bounded
 * path: success against a real local listener, a distinct (non-success,
 * non-hanging) result against a refused/unreachable target, and that EVERY
 * call returns far below the OS multi-second default — which is exactly the
 * guarantee that protects the delaymgr heartbeat thread (R2.3).
 */

#define _GNU_SOURCE
#include <arpa/inet.h>
#include <errno.h>
#include <netinet/in.h>
#include <stdio.h>
#include <string.h>
#include <sys/socket.h>
#include <time.h>
#include <unistd.h>

#include <unity/unity.h>

#include "config.h"
#include "netio.h"

/* Generous upper bound: well below the multi-second (typically 20-130s) OS
 * connect timeout, generous above the configured DEFAULT_CONNECT_TIMEOUT_MS so
 * the assertion is robust on a loaded CI box. The point is "bounded, not the OS
 * default", not a tight latency SLA. */
#define BOUND_CEIL_MS 5000.0

static double now_ms(void)
{
    struct timespec t;
    clock_gettime(CLOCK_MONOTONIC, &t);
    return (double)t.tv_sec * 1000.0 + (double)t.tv_nsec / 1e6;
}

/* Open a listening TCP socket on 127.0.0.1, return its fd and the chosen port
 * (ephemeral) via *port_out. */
static int open_local_listener(int *port_out)
{
    int lfd = socket(AF_INET, SOCK_STREAM, 0);
    TEST_ASSERT_TRUE(lfd >= 0);

    struct sockaddr_in a;
    memset(&a, 0, sizeof(a));
    a.sin_family      = AF_INET;
    a.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
    a.sin_port        = 0; /* ephemeral */

    TEST_ASSERT_EQUAL_INT(0, bind(lfd, (struct sockaddr *)&a, sizeof(a)));
    TEST_ASSERT_EQUAL_INT(0, listen(lfd, 8));

    socklen_t alen = sizeof(a);
    TEST_ASSERT_EQUAL_INT(0, getsockname(lfd, (struct sockaddr *)&a, &alen));
    *port_out = (int)ntohs(a.sin_port);
    return lfd;
}

/* ── Test 1: connect to a reachable local listener succeeds + usable fd ── */
static void test_bounded_connect_reachable_succeeds(void)
{
    int  port = 0;
    int  lfd  = open_local_listener(&port);

    int    cfd = socket(AF_INET, SOCK_STREAM, 0);
    TEST_ASSERT_TRUE(cfd >= 0);

    char host[] = "127.0.0.1";
    double t0  = now_ms();
    int    rc  = socket_connect(cfd, host, port, 500);
    double dt  = now_ms() - t0;

    TEST_ASSERT_TRUE_MESSAGE(rc >= 0, "connect to reachable local listener must succeed");
    TEST_ASSERT_TRUE_MESSAGE(dt < BOUND_CEIL_MS, "reachable connect must return promptly");

    /* The fd must be usable: the listener side must see a pending connection. */
    int afd = accept(lfd, NULL, NULL);
    TEST_ASSERT_TRUE_MESSAGE(afd >= 0, "listener must accept the connected fd");

    if (afd >= 0)
        close(afd);
    close(cfd);
    close(lfd);
    printf("test_bounded_connect_reachable_succeeds: PASS (%.0fms)\n", dt);
}

/* ── Test 2: connect to a refused (closed) port returns a distinct error,
 *           bounded, never a success, never a multi-second hang. ── */
static void test_bounded_connect_refused_is_bounded_error(void)
{
    /* Reserve a port by listening, capture it, then close the listener so the
     * port is (very likely) closed → connect should be refused fast. */
    int port = 0;
    int lfd  = open_local_listener(&port);
    close(lfd);

    int cfd = socket(AF_INET, SOCK_STREAM, 0);
    TEST_ASSERT_TRUE(cfd >= 0);

    char host[] = "127.0.0.1";
    double t0  = now_ms();
    int    rc  = socket_connect(cfd, host, port, 500);
    double dt  = now_ms() - t0;

    /* Whatever the verdict (refused / timeout), it MUST be bounded: this is the
     * core R2.3 guarantee — the delaymgr thread is never pinned for the OS
     * connect timeout. */
    TEST_ASSERT_TRUE_MESSAGE(dt < BOUND_CEIL_MS,
        "refused/unreachable connect must be bounded, not the OS default");

    close(cfd);
    printf("test_bounded_connect_refused_is_bounded_error: PASS (rc=%d, %.0fms)\n", rc, dt);
}

/* ── Test 3: timeout==0 resolves to DEFAULT_CONNECT_TIMEOUT_MS and the call
 *           still returns bounded (default is honored, not infinite). ── */
static void test_bounded_connect_zero_uses_default(void)
{
    int  port = 0;
    int  lfd  = open_local_listener(&port);

    int    cfd = socket(AF_INET, SOCK_STREAM, 0);
    TEST_ASSERT_TRUE(cfd >= 0);

    char host[] = "127.0.0.1";
    double t0  = now_ms();
    int    rc  = socket_connect(cfd, host, port, 0); /* 0 => default */
    double dt  = now_ms() - t0;

    TEST_ASSERT_TRUE_MESSAGE(rc >= 0, "reachable connect with default timeout must succeed");
    /* The resolved bound is DEFAULT_CONNECT_TIMEOUT_MS; the call must return
     * within a generous multiple of it (never the OS default). */
    TEST_ASSERT_TRUE_MESSAGE(dt < BOUND_CEIL_MS, "default-timeout connect must be bounded");

    int afd = accept(lfd, NULL, NULL);
    TEST_ASSERT_TRUE(afd >= 0);
    if (afd >= 0)
        close(afd);
    close(cfd);
    close(lfd);
    printf("test_bounded_connect_zero_uses_default: PASS (default=%dms, %.0fms)\n",
           DEFAULT_CONNECT_TIMEOUT_MS, dt);
}

/* ── Test suite entry point ───────────────────────────────────────────── */
void test_netio_bounded_connect(void)
{
    printf("\n>> netio bounded-connect (R2.3) tests\n\n");

    RUN_TEST(test_bounded_connect_reachable_succeeds);
    RUN_TEST(test_bounded_connect_refused_is_bounded_error);
    RUN_TEST(test_bounded_connect_zero_uses_default);
}