#define _GNU_SOURCE
#include <dirent.h>
#include <errno.h>
#include <fcntl.h>
#include <limits.h>
#include <poll.h>
#include <signal.h>
#include <stdarg.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/inotify.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <sys/wait.h>
#include <time.h>
#include <unistd.h>

enum
{
	Failtimeout = 8000,
};

typedef struct Test Test;

struct Test
{
	pid_t child;
	int errfd;
	int notifyfd;
	int runtimewd;
	int buswd;
	int endpointclean;
	char root[256];
	char runtime[320];
	char config[320];
	char ibus[384];
	char bus[448];
	char home[320];
	char socket[384];
	char badmap[320];
	char expected[768];
	char err[4096];
	size_t nerr;
};

static uint32_t endpointmask = IN_CREATE|IN_MOVED_TO|IN_MOVED_FROM|IN_ATTRIB|
	IN_DELETE|IN_MODIFY|IN_CLOSE_WRITE|IN_DELETE_SELF|IN_MOVE_SELF;

static int
fail(char *fmt, ...)
{
	va_list ap;

	fprintf(stderr, "daemon_failure_test: ");
	va_start(ap, fmt);
	vfprintf(stderr, fmt, ap);
	va_end(ap);
	fputc('\n', stderr);
	return 0;
}

static int64_t
nowms(void)
{
	struct timespec ts;

	if(clock_gettime(CLOCK_MONOTONIC, &ts) < 0)
		return -1;
	return (int64_t)ts.tv_sec * 1000 + ts.tv_nsec / 1000000;
}

static int
leftms(int64_t deadline)
{
	int64_t n;

	n = nowms();
	if(n < 0)
		return -1;
	n = deadline - n;
	if(n <= 0)
		return 0;
	if(n > 0x7fffffff)
		return 0x7fffffff;
	return n;
}

static int
makedir(char *path)
{
	if(mkdir(path, 0700) == 0)
		return 1;
	return fail("mkdir %s: %s", path, strerror(errno));
}

static void
readerrors(Test *t)
{
	ssize_t n;

	if(t->errfd < 0)
		return;
	while(t->nerr + 1 < sizeof t->err){
		n = read(t->errfd, t->err + t->nerr,
			sizeof t->err - t->nerr - 1);
		if(n > 0){
			t->nerr += n;
			continue;
		}
		if(n < 0 && errno == EINTR)
			continue;
		break;
	}
	t->err[t->nerr] = '\0';
}

static void
showerrors(Test *t)
{
	readerrors(t);
	if(t->nerr != 0)
		fprintf(stderr, "daemon_failure_test: daemon stderr:\n%s", t->err);
}

static int
setup(Test *t)
{
	memset(t, 0, sizeof *t);
	t->child = -1;
	t->errfd = -1;
	t->notifyfd = -1;
	t->runtimewd = -1;
	t->buswd = -1;
	t->endpointclean = 1;
	if(nowms() < 0)
		return fail("read monotonic clock: %s", strerror(errno));
	snprintf(t->root, sizeof t->root, "/tmp/strans-failure.XXXXXX");
	if(mkdtemp(t->root) == NULL){
		t->root[0] = '\0';
		return fail("mkdtemp: %s", strerror(errno));
	}
	if(snprintf(t->runtime, sizeof t->runtime, "%s/runtime", t->root)
	   >= (int)sizeof t->runtime ||
	   snprintf(t->config, sizeof t->config, "%s/config", t->root)
	   >= (int)sizeof t->config ||
	   snprintf(t->ibus, sizeof t->ibus, "%s/ibus", t->config)
	   >= (int)sizeof t->ibus ||
	   snprintf(t->bus, sizeof t->bus, "%s/bus", t->ibus)
	   >= (int)sizeof t->bus ||
	   snprintf(t->home, sizeof t->home, "%s/home", t->root)
	   >= (int)sizeof t->home ||
	   snprintf(t->socket, sizeof t->socket, "%s/strans.sock", t->runtime)
	   >= (int)sizeof t->socket ||
	   snprintf(t->badmap, sizeof t->badmap, "%s/missing-map", t->root)
	   >= (int)sizeof t->badmap ||
	   snprintf(t->expected, sizeof t->expected,
	   "strans: can't open: %s/hira.map\n", t->badmap)
	   >= (int)sizeof t->expected)
		return fail("temporary path is too long");
	if(!makedir(t->runtime) || !makedir(t->config) || !makedir(t->ibus) ||
	   !makedir(t->bus) || !makedir(t->home))
		return 0;
	t->notifyfd = inotify_init1(IN_CLOEXEC|IN_NONBLOCK);
	if(t->notifyfd < 0)
		return fail("inotify_init1: %s", strerror(errno));
	t->runtimewd = inotify_add_watch(t->notifyfd, t->runtime, endpointmask);
	if(t->runtimewd < 0)
		return fail("watch runtime directory: %s", strerror(errno));
	t->buswd = inotify_add_watch(t->notifyfd, t->bus, endpointmask);
	if(t->buswd < 0)
		return fail("watch IBus directory: %s", strerror(errno));
	return 1;
}

static int
startchild(Test *t, char *program)
{
	int errpipe[2], fd;
	long maxfd;
	pid_t pid;

	if(pipe2(errpipe, O_CLOEXEC|O_NONBLOCK) < 0)
		return fail("pipe2: %s", strerror(errno));
	pid = fork();
	if(pid < 0){
		close(errpipe[0]);
		close(errpipe[1]);
		return fail("fork: %s", strerror(errno));
	}
	if(pid == 0){
		close(errpipe[0]);
		if(dup2(errpipe[1], STDERR_FILENO) < 0)
			_exit(126);
		close(errpipe[1]);
		if(close_range(3, UINT_MAX, 0) < 0){
			maxfd = sysconf(_SC_OPEN_MAX);
			if(maxfd < 0)
				maxfd = 1024;
			for(fd = 3; fd < maxfd; fd++)
				close(fd);
		}
		if(setenv("XDG_RUNTIME_DIR", t->runtime, 1) < 0 ||
		   setenv("XDG_CONFIG_HOME", t->config, 1) < 0 ||
		   setenv("HOME", t->home, 1) < 0 || unsetenv("DISPLAY") < 0 ||
		   unsetenv("DBUS_SESSION_BUS_ADDRESS") < 0 ||
		   unsetenv("IBUS_ADDRESS") < 0){
			dprintf(STDERR_FILENO, "set daemon environment: %s\n",
				strerror(errno));
			_exit(126);
		}
		execl(program, program, t->badmap, (char*)0);
		dprintf(STDERR_FILENO, "exec %s: %s\n", program, strerror(errno));
		_exit(127);
	}
	close(errpipe[1]);
	t->child = pid;
	t->errfd = errpipe[0];
	return 1;
}

static int
drainnotify(Test *t)
{
	char buf[4096];
	char *where;
	struct inotify_event *ev;
	ssize_t n;
	size_t off;
	int ok;

	ok = 1;
	for(;;){
		n = read(t->notifyfd, buf, sizeof buf);
		if(n < 0 && errno == EINTR)
			continue;
		if(n < 0 && errno == EAGAIN)
			return ok;
		if(n < 0)
			return fail("read endpoint watches: %s", strerror(errno));
		if(n == 0)
			return ok;
		for(off = 0; off + sizeof *ev <= (size_t)n;
		   off += sizeof *ev + ev->len){
			ev = (struct inotify_event*)(buf + off);
			if(off + sizeof *ev + ev->len > (size_t)n)
				return fail("truncated inotify event");
			if(ev->mask & IN_Q_OVERFLOW){
				t->endpointclean = 0;
				fail("endpoint watch queue overflowed");
				ok = 0;
				continue;
			}
			if(ev->mask & (IN_IGNORED|IN_UNMOUNT)){
				t->endpointclean = 0;
				fail("endpoint watch became invalid: %#x", ev->mask);
				ok = 0;
				continue;
			}
			if(ev->wd != t->runtimewd && ev->wd != t->buswd)
				continue;
			if(!(ev->mask & endpointmask))
				continue;
			where = ev->wd == t->runtimewd ? "runtime" : "IBus";
			t->endpointclean = 0;
			fail("unexpected %s endpoint event %#x for %s", where,
				ev->mask, ev->len != 0 ? ev->name : "directory");
			ok = 0;
		}
		if(off != (size_t)n)
			return fail("truncated inotify event buffer");
	}
}

static int
killowned(Test *t, int *status)
{
	pid_t pid;
	int n, ok;

	if(t->child <= 0)
		return 1;
	pid = t->child;
	ok = 1;
	if(kill(pid, SIGKILL) < 0 && errno != ESRCH){
		fail("kill -9 %ld: %s", (long)pid, strerror(errno));
		ok = 0;
	}
	do
		n = waitpid(pid, status, 0);
	while(n < 0 && errno == EINTR);
	if(n != pid){
		fail("reap %ld after SIGKILL: %s", (long)pid,
			n < 0 ? strerror(errno) : "wrong child");
		ok = 0;
	}else
		t->child = -1;
	return ok;
}

static int
waitfailed(Test *t)
{
	struct pollfd pfd[2];
	pid_t pid;
	int n, ok, status, timeout;
	int64_t deadline;

	pid = t->child;
	ok = 1;
	deadline = nowms();
	if(deadline < 0){
		fail("read monotonic clock before waiting for daemon: %s",
			strerror(errno));
		killowned(t, &status);
		readerrors(t);
		drainnotify(t);
		return 0;
	}
	deadline += Failtimeout;
	for(;;){
		do
			n = waitpid(pid, &status, WNOHANG);
		while(n < 0 && errno == EINTR);
		if(n == pid){
			t->child = -1;
			readerrors(t);
			if(!drainnotify(t))
				ok = 0;
			break;
		}
		if(n < 0){
			fail("waitpid failed daemon %ld: %s", (long)pid,
				strerror(errno));
			if(errno == ECHILD){
				t->child = -1;
				readerrors(t);
				drainnotify(t);
			}else{
				killowned(t, &status);
				readerrors(t);
				drainnotify(t);
			}
			return 0;
		}
		timeout = leftms(deadline);
		if(timeout < 0){
			fail("read monotonic clock while waiting for daemon: %s",
				strerror(errno));
			killowned(t, &status);
			readerrors(t);
			drainnotify(t);
			return 0;
		}
		if(timeout == 0){
			fail("daemon %ld did not exit after map initialization failure",
				(long)pid);
			killowned(t, &status);
			return 0;
		}
		pfd[0].fd = t->notifyfd;
		pfd[0].events = POLLIN;
		pfd[0].revents = 0;
		pfd[1].fd = t->errfd;
		pfd[1].events = POLLIN|POLLHUP;
		pfd[1].revents = 0;
		n = poll(pfd, 2, timeout);
		if(n < 0 && errno == EINTR)
			continue;
		if(n < 0){
			fail("poll failed daemon: %s", strerror(errno));
			killowned(t, &status);
			readerrors(t);
			drainnotify(t);
			return 0;
		}
		if(pfd[0].revents & (POLLERR|POLLHUP|POLLNVAL)){
			fail("endpoint watch became unusable: %#x", pfd[0].revents);
			killowned(t, &status);
			readerrors(t);
			drainnotify(t);
			return 0;
		}
		if(pfd[1].revents & (POLLERR|POLLNVAL)){
			fail("daemon stderr pipe became unusable: %#x",
				pfd[1].revents);
			killowned(t, &status);
			readerrors(t);
			drainnotify(t);
			return 0;
		}
		if((pfd[0].revents & POLLIN) && !drainnotify(t))
			ok = 0;
		if(pfd[1].revents & (POLLIN|POLLHUP))
			readerrors(t);
	}
	if(!WIFEXITED(status) || WEXITSTATUS(status) != 1)
		return fail("daemon map failure had wait status %#x", status);
	if(strcmp(t->err, t->expected) != 0)
		return fail("daemon did not report the expected map initialization failure");
	return ok && t->endpointclean;
}

static int
dirempty(char *path, char *which)
{
	DIR *dir;
	struct dirent *de;
	int count, ok;

	dir = opendir(path);
	if(dir == NULL)
		return fail("open %s directory %s: %s", which, path, strerror(errno));
	count = 0;
	ok = 1;
	errno = 0;
	while((de = readdir(dir)) != NULL){
		if(strcmp(de->d_name, ".") == 0 || strcmp(de->d_name, "..") == 0)
			continue;
		count++;
		fail("unexpected %s directory entry %s", which, de->d_name);
		ok = 0;
		errno = 0;
	}
	if(errno != 0){
		fail("read %s directory %s: %s", which, path, strerror(errno));
		ok = 0;
	}
	if(closedir(dir) < 0){
		fail("close %s directory %s: %s", which, path, strerror(errno));
		ok = 0;
	}
	if(count != 0)
		ok = 0;
	return ok;
}

static int
checkendpoints(Test *t)
{
	struct stat st;
	int ok;

	ok = 1;
	errno = 0;
	if(lstat(t->socket, &st) == 0){
		fail("IPC endpoint exists after failed startup: %s", t->socket);
		ok = 0;
	}else if(errno != ENOENT){
		fail("lstat failed IPC endpoint: %s", strerror(errno));
		ok = 0;
	}
	if(!dirempty(t->runtime, "runtime"))
		ok = 0;
	if(!dirempty(t->bus, "IBus"))
		ok = 0;
	return ok && t->endpointclean;
}

static int clearfd(int);

static int
removeentry(int dirfd, char *name)
{
	struct stat st;
	int fd, ok;

	if(fstatat(dirfd, name, &st, AT_SYMLINK_NOFOLLOW) < 0)
		return fail("stat cleanup entry %s: %s", name, strerror(errno));
	if(!S_ISDIR(st.st_mode)){
		if(unlinkat(dirfd, name, 0) == 0 || errno == ENOENT)
			return 1;
		return fail("remove cleanup entry %s: %s", name, strerror(errno));
	}
	fd = openat(dirfd, name, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOFOLLOW);
	if(fd < 0)
		return fail("open cleanup directory %s: %s", name, strerror(errno));
	ok = clearfd(fd);
	if(close(fd) < 0){
		fail("close cleanup directory %s: %s", name, strerror(errno));
		ok = 0;
	}
	if(unlinkat(dirfd, name, AT_REMOVEDIR) < 0 && errno != ENOENT){
		fail("remove cleanup directory %s: %s", name, strerror(errno));
		ok = 0;
	}
	return ok;
}

static int
clearfd(int fd)
{
	DIR *dir;
	struct dirent *de;
	int copy, ok;

	copy = dup(fd);
	if(copy < 0)
		return fail("duplicate cleanup directory: %s", strerror(errno));
	dir = fdopendir(copy);
	if(dir == NULL){
		close(copy);
		return fail("open cleanup directory stream: %s", strerror(errno));
	}
	ok = 1;
	errno = 0;
	while((de = readdir(dir)) != NULL){
		if(strcmp(de->d_name, ".") == 0 || strcmp(de->d_name, "..") == 0)
			continue;
		if(!removeentry(fd, de->d_name))
			ok = 0;
		errno = 0;
	}
	if(errno != 0){
		fail("read cleanup directory: %s", strerror(errno));
		ok = 0;
	}
	if(closedir(dir) < 0){
		fail("close cleanup directory stream: %s", strerror(errno));
		ok = 0;
	}
	return ok;
}

static int
removeroot(Test *t)
{
	int fd, ok;

	if(t->root[0] == '\0')
		return 1;
	fd = open(t->root, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOFOLLOW);
	if(fd < 0)
		return errno == ENOENT || fail("open cleanup root %s: %s", t->root,
			strerror(errno));
	ok = clearfd(fd);
	if(close(fd) < 0){
		fail("close cleanup root %s: %s", t->root, strerror(errno));
		ok = 0;
	}
	if(rmdir(t->root) < 0 && errno != ENOENT){
		fail("rmdir %s: %s", t->root, strerror(errno));
		ok = 0;
	}
	return ok;
}

static int
cleanup(Test *t)
{
	int n, ok, status;

	ok = 1;
	if(t->child > 0){
		do
			n = waitpid(t->child, &status, WNOHANG);
		while(n < 0 && errno == EINTR);
		if(n == t->child)
			t->child = -1;
		else if(n == 0){
			fail("failed-start daemon still running during cleanup");
			ok = 0;
			if(!killowned(t, &status))
				ok = 0;
		}else{
			fail("check failed-start daemon during cleanup: %s",
				strerror(errno));
			ok = 0;
			if(errno == ECHILD)
				t->child = -1;
			else if(!killowned(t, &status))
				ok = 0;
		}
	}
	readerrors(t);
	if(t->errfd >= 0){
		if(close(t->errfd) < 0){
			fail("close daemon stderr: %s", strerror(errno));
			ok = 0;
		}
		t->errfd = -1;
	}
	if(t->notifyfd >= 0){
		if(t->runtimewd >= 0 &&
		   inotify_rm_watch(t->notifyfd, t->runtimewd) < 0 && errno != EINVAL){
			fail("remove runtime watch: %s", strerror(errno));
			ok = 0;
		}
		if(t->buswd >= 0 &&
		   inotify_rm_watch(t->notifyfd, t->buswd) < 0 && errno != EINVAL){
			fail("remove IBus watch: %s", strerror(errno));
			ok = 0;
		}
		if(close(t->notifyfd) < 0){
			fail("close endpoint watches: %s", strerror(errno));
			ok = 0;
		}
		t->notifyfd = -1;
	}
	if(!removeroot(t))
		ok = 0;
	return ok;
}

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

	if(argc != 2){
		fprintf(stderr, "usage: daemon_failure_test strans\n");
		return 2;
	}
	ok = setup(&test);
	if(ok)
		ok = startchild(&test, argv[1]);
	if(ok){
		ok = waitfailed(&test);
		if(test.child <= 0 && !checkendpoints(&test))
			ok = 0;
	}
	if(!cleanup(&test))
		ok = 0;
	if(!ok){
		showerrors(&test);
		return 1;
	}
	printf("failed daemon startup published no endpoints: ok\n");
	return 0;
}