strans/tests/daemon_restart_test.c

#define _GNU_SOURCE
#include <errno.h>
#include <poll.h>
#include <signal.h>
#include <stdint.h>
#include <stdio.h>
#include <string.h>
#include <sys/socket.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <sys/wait.h>
#include <unistd.h>

#include "ipc.h"
#include "live.h"
#include "livebus.h"

enum
{
	Ctrlmask = 1<<2,
};

typedef struct Test Test;

struct Test
{
	Live l;
	Daemon first;
	Daemon second;
	pid_t firstpid;
	int ipcfirst;
	int ipcnew;
	int busfirstfd;
	DBusConnection *busfirst;
	DBusConnection *busnew;
	char firstaddress[512];
	char secondaddress[512];
};

static int
setup(Test *t)
{
	struct sigaction sa;

	memset(t, 0, sizeof *t);
	daemoninit(&t->first, "first daemon");
	daemoninit(&t->second, "second daemon");
	t->firstpid = -1;
	t->ipcfirst = -1;
	t->ipcnew = -1;
	t->busfirstfd = -1;
	memset(&sa, 0, sizeof sa);
	sa.sa_handler = SIG_DFL;
	if(sigemptyset(&sa.sa_mask) < 0 || sigaction(SIGCHLD, &sa, NULL) < 0)
		return fail("establish exact child ownership: %s", strerror(errno));
	return livesetup(&t->l, "restart");
}

static int
privateaddress(char *address, pid_t pid)
{
	char prefix[96];
	size_t n;

	if(snprintf(prefix, sizeof prefix, "unix:abstract=strans-%ld",
	   (long)pid) >= (int)sizeof prefix)
		return fail("private IBus prefix is too long");
	n = strlen(prefix);
	return (strncmp(address, prefix, n) == 0 && address[n] == ',') ||
		fail("IBus address is not the daemon's private abstract address");
}

static int
focusin(DBusConnection *conn, char *path)
{
	DBusMessage *reply;

	reply = callret(conn, contextcall(path, "FocusIn"),
		"replacement FocusIn");
	if(reply == NULL)
		return 0;
	if(!dbus_message_has_signature(reply, "")){
		dbus_message_unref(reply);
		return fail("replacement FocusIn returned a nonempty reply");
	}
	dbus_message_unref(reply);
	return 1;
}

static int
keycall(DBusConnection *conn, char *path, dbus_uint32_t sym,
	dbus_uint32_t state, int expected, char *where)
{
	DBusMessage *m, *reply;
	DBusError err;
	dbus_uint32_t code;
	dbus_bool_t eaten;

	code = 0;
	m = contextcall(path, "ProcessKeyEvent");
	if(m == NULL || !dbus_message_append_args(m,
	   DBUS_TYPE_UINT32, &sym, DBUS_TYPE_UINT32, &code,
	   DBUS_TYPE_UINT32, &state, DBUS_TYPE_INVALID)){
		if(m != NULL)
			dbus_message_unref(m);
		return fail("build %s ProcessKeyEvent call", where);
	}
	reply = callret(conn, m, where);
	if(reply == NULL)
		return 0;
	dbus_error_init(&err);
	if(!dbus_message_has_signature(reply, "b") ||
	   !dbus_message_get_args(reply, &err, DBUS_TYPE_BOOLEAN, &eaten,
	   DBUS_TYPE_INVALID)){
		dbus_message_unref(reply);
		dbus_error_free(&err);
		return fail("%s returned an invalid ProcessKeyEvent reply", where);
	}
	dbus_error_free(&err);
	dbus_message_unref(reply);
	return (eaten != FALSE) == (expected != 0) ||
		fail("%s eaten=%d, expected %d", where, eaten != FALSE, expected);
}

static int
openpersistent(Test *t)
{
	t->ipcfirst = connectsocket(t->l.socket, nowms() + Calltimeout);
	if(t->ipcfirst < 0)
		return fail("connect persistent IPC client: %s", strerror(errno));
	if(!ipcprobe(t->ipcfirst, "persistent pre-crash"))
		return 0;
	t->busfirst = openbus(t->firstaddress, "persistent pre-crash");
	if(t->busfirst == NULL ||
	   !hello(t->busfirst, NULL, 0, "persistent pre-crash Hello") ||
	   !createcontext(t->busfirst, NULL, 0, "persistent pre-crash context"))
		return 0;
	if(!dbus_connection_get_unix_fd(t->busfirst, &t->busfirstfd))
		return fail("persistent IBus connection has no Unix descriptor");
	return 1;
}

/*
 * plan9port ignores a broken pipe: the note handlers run and the daemon
 * goes on serving, so the endpoints it announced must still be there.
 */
static int
survivenote(Test *t)
{
	struct stat st;
	int fd, ok;

	if(kill(t->firstpid, SIGPIPE) < 0)
		return fail("send a broken pipe to the first daemon: %s",
			strerror(errno));
	pausems(200);
	if(!daemonalive(&t->first))
		return fail("first daemon died of a broken pipe");
	if(lstat(t->l.socket, &st) < 0)
		return fail("IPC socket taken by a note the daemon survived: %s",
			strerror(errno));
	if(lstat(t->l.addrfile, &st) < 0)
		return fail("IBus address taken by a note the daemon survived: %s",
			strerror(errno));
	fd = connectsocket(t->l.socket, nowms() + Calltimeout);
	if(fd < 0)
		return fail("connect after a broken pipe: %s", strerror(errno));
	ok = ipcprobe(fd, "after a broken pipe");
	if(close(fd) < 0)
		return fail("close the client opened after a broken pipe: %s",
			strerror(errno));
	return ok;
}

static int
hardcrash(Test *t)
{
	int status;

	status = 0;
	if(!daemonalive(&t->first) || !killdaemon(&t->first, &status))
		return 0;
	readerrors(&t->first);
	if(!WIFSIGNALED(status) || WTERMSIG(status) != SIGKILL)
		return fail("first daemon hard crash had wait status %#x", status);
	return 1;
}

static int
waitipcclosed(Test *t)
{
	struct pollfd pfd;
	unsigned char byte;
	ssize_t n;
	int timeout;
	int64_t deadline;

	deadline = nowms() + Calltimeout;
	for(;;){
		timeout = leftms(deadline);
		if(timeout == 0)
			return fail("persistent IPC connection did not disconnect promptly");
		pfd.fd = t->ipcfirst;
		pfd.events = POLLIN|POLLHUP|POLLERR;
		pfd.revents = 0;
		n = poll(&pfd, 1, timeout);
		if(n < 0 && errno == EINTR)
			continue;
		if(n <= 0)
			return fail("poll persistent IPC disconnection: %s",
				n == 0 ? "timed out" : strerror(errno));
		n = recv(t->ipcfirst, &byte, 1, MSG_PEEK);
		if(n == 0 || (n < 0 && (errno == ECONNRESET || errno == ENOTCONN ||
		   errno == EPIPE)))
			return 1;
		if(n < 0 && (errno == EINTR || errno == EAGAIN || errno == EWOULDBLOCK))
			continue;
		if(n < 0)
			return fail("read persistent IPC disconnection: %s",
				strerror(errno));
		return fail("persistent IPC connection retained unread data after crash");
	}
}

static int
waitbusclosed(Test *t)
{
	struct pollfd pfd;
	int n, timeout;
	int64_t deadline;

	deadline = nowms() + Calltimeout;
	for(;;){
		if(!dbus_connection_get_is_connected(t->busfirst))
			return 1;
		timeout = leftms(deadline);
		if(timeout == 0)
			return fail("persistent IBus connection did not disconnect promptly");
		pfd.fd = t->busfirstfd;
		pfd.events = POLLIN|POLLHUP|POLLERR;
		pfd.revents = 0;
		n = poll(&pfd, 1, timeout);
		if(n < 0 && errno == EINTR)
			continue;
		if(n <= 0)
			return fail("poll persistent IBus disconnection: %s",
				n == 0 ? "timed out" : strerror(errno));
		if(pfd.revents & POLLNVAL)
			return fail("persistent IBus descriptor became invalid");
		dbus_connection_read_write_dispatch(t->busfirst, 0);
	}
}

static int
checkstale(Test *t)
{
	struct stat st;
	char contents[2048], address[512];
	size_t ncontents;
	pid_t declared;

	if(lstat(t->l.socket, &st) < 0)
		return fail("stale IPC socket disappeared: %s", strerror(errno));
	if(!S_ISSOCK(st.st_mode))
		return fail("stale IPC endpoint is not a socket");
	if(lstat(t->l.addrfile, &st) < 0)
		return fail("stale IBus address file disappeared: %s", strerror(errno));
	if(!readfile(t->l.addrfile, contents, sizeof contents, &ncontents) ||
	   !parseaddress(contents, ncontents, address, sizeof address, &declared))
		return 0;
	if(declared != t->firstpid || strcmp(address, t->firstaddress) != 0)
		return fail("stale IBus address no longer names the dead first daemon");
	return 1;
}

/* The dead daemon's abstract socket has no listener, so this must fail. */
static int
rejectold(Test *t, char *where)
{
	DBusConnection *conn;
	DBusError err;
	int ok;

	dbus_error_init(&err);
	conn = dbus_connection_open_private(t->firstaddress, &err);
	if(conn != NULL){
		dbus_connection_set_exit_on_disconnect(conn, FALSE);
		closebus(&conn);
		ok = fail("old private address accepted a connection %s", where);
	}else
		ok = dbus_error_is_set(&err) ||
			fail("old private address failed without a D-Bus error %s", where);
	dbus_error_free(&err);
	return ok;
}

static int
openreplacement(Test *t)
{
	unsigned char selectjp[] = {1, 0, 0, 0, 0, 0};
	unsigned char keyk[] = {1, 0, 0, 0, 1, 0, 'k'};
	char path[96];

	t->ipcnew = connectsocket(t->l.socket, nowms() + Calltimeout);
	if(t->ipcnew < 0)
		return fail("connect replacement IPC client: %s", strerror(errno));
	if(!ipcrequest(t->ipcnew, Mctrl, 'n', selectjp, sizeof selectjp,
	   "replacement select Japanese") ||
	   !ipcrequest(t->ipcnew, 0, 'k', keyk, sizeof keyk,
	   "replacement real key"))
		return 0;
	t->busnew = openbus(t->secondaddress, "replacement");
	if(t->busnew == NULL ||
	   !hello(t->busnew, NULL, 0, "replacement Hello") ||
	   !createcontext(t->busnew, path, sizeof path, "replacement context") ||
	   !focusin(t->busnew, path) ||
	   !keycall(t->busnew, path, 'n', Ctrlmask, 1,
	   "replacement select Japanese") ||
	   !keycall(t->busnew, path, 'k', 0, 1, "replacement real key"))
		return 0;
	return daemonalive(&t->second);
}

static int
cleanup(Test *t)
{
	int ok;

	ok = 1;
	closebus(&t->busnew);
	closebus(&t->busfirst);
	if(t->ipcnew >= 0){
		if(close(t->ipcnew) < 0)
			ok = fail("close replacement IPC client: %s", strerror(errno));
		t->ipcnew = -1;
	}
	if(t->ipcfirst >= 0){
		if(close(t->ipcfirst) < 0)
			ok = fail("close persistent IPC client: %s", strerror(errno));
		t->ipcfirst = -1;
	}
	if(!killdaemon(&t->first, NULL))
		ok = 0;
	if(!stopdaemon(&t->second))
		ok = 0;
	if(!closeerrors(&t->first))
		ok = 0;
	if(!closeerrors(&t->second))
		ok = 0;
	if(!liveclean(&t->l))
		ok = 0;
	return ok;
}

static int
runrestart(Test *t, char *program, char *mapdir)
{
	if(!startdaemon(&t->l, &t->first, program, mapdir))
		return 0;
	t->firstpid = t->first.pid;
	if(!waitready(&t->l, &t->first, t->firstaddress, sizeof t->firstaddress) ||
	   !privateaddress(t->firstaddress, t->firstpid) || !openpersistent(t))
		return 0;
	if(!survivenote(t))
		return 0;
	if(!hardcrash(t) || !waitipcclosed(t) || !waitbusclosed(t) ||
	   !checkstale(t) || !rejectold(t, "after hard crash"))
		return 0;
	closebus(&t->busfirst);
	if(close(t->ipcfirst) < 0)
		return fail("close disconnected persistent IPC client: %s",
			strerror(errno));
	t->ipcfirst = -1;
	if(!startdaemon(&t->l, &t->second, program, mapdir) ||
	   !waitready(&t->l, &t->second, t->secondaddress,
	   sizeof t->secondaddress) ||
	   !privateaddress(t->secondaddress, t->second.pid) ||
	   !openreplacement(t))
		return 0;
	return daemonalive(&t->second);
}

int
main(int argc, char **argv)
{
	Test test;
	int ok;

	testname = "daemon_restart_test";
	if(argc != 3){
		fprintf(stderr, "usage: daemon_restart_test strans mapdir\n");
		return 2;
	}
	ok = setup(&test);
	if(ok)
		ok = runrestart(&test, argv[1], argv[2]);
	if(!cleanup(&test))
		ok = 0;
	if(!ok){
		showerrors(&test.first);
		showerrors(&test.second);
		return 1;
	}
	printf("daemon endpoints across a note and a hard crash: ok\n");
	return 0;
}