// SPDX-License-Identifier: GPL-2.0-or-later
/*
 * Copyright (c) 2017 Richard Palethorpe <rpalethorpe@suse.com>
 */
/*
 * CVE-2016-7117
 *
 * This tests for a use after free caused by a race between recvmmsg() and
 * close(). The exit path for recvmmsg() in (a2e2725541f: net: Introduce
 * recvmmsg socket syscall) called fput() on the active file descriptor before
 * checking the error state and setting the socket's error field.
 *
 * If one or more messages are received by recvmmsg() followed by one which
 * fails, the socket's error field will be set. If just after recvmmsg() calls
 * fput(), a call to close() is made on the same file descriptor there is a
 * race between close() releasing the socket object and recvmmsg() setting its
 * error field.
 *
 * fput() does not release a file descriptor's resources (e.g. a socket)
 * immediatly, it queues them to be released just before a system call returns
 * to user land. So the close() system call must call fput() after it is
 * called in recvmmsg(), exit and release the resources all before the socket
 * error is set.
 *
 * Usually if the vulnerability is present the test will be killed with a
 * kernel null pointer exception. However this is not guaranteed to happen
 * every time.
 *
 * The following was used for reference
 * https://blog.lizzie.io/notes-about-cve-2016-7117.html
 */

#include "config.h"

#include <sys/wait.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <sys/syscall.h>
#include <stdlib.h>
#include <errno.h>

#include "tst_test.h"
#include "tst_safe_net.h"
#include "tst_timer.h"
#include "tst_fuzzy_sync.h"

/* The bug was present in the kernel before recvmmsg was exposed by glibc */
#include "lapi/syscalls.h"
#include "lapi/socket.h"

#define MSG "abcdefghijklmnop"
#define RECV_TIMEOUT 1
#define ATTEMPTS 0x1FFFFF

static volatile int socket_fds[2];
static struct mmsghdr msghdrs[2] = {
	{
		.msg_hdr = {
			.msg_iov = &(struct iovec) {
				.iov_len = sizeof(MSG),
			},
			.msg_iovlen = 1
		}
	},
	{
		.msg_hdr = {
			.msg_iov = &(struct iovec) {
				.iov_base = (void *)(0xbadadd),
				.iov_len = ~0,
			},
			.msg_iovlen = 1
		}
	}
};
static char rbuf[sizeof(MSG)];
static struct timespec timeout = { .tv_sec = RECV_TIMEOUT };
static struct tst_fzsync_pair fzsync_pair;
static void *send_and_close(void *);

static void setup(void)
{
	fzsync_pair.min_samples = 10000;

	tst_syscall(__NR_recvmmsg, 0, 0, 0, 0, 0);

	tst_fzsync_pair_init(&fzsync_pair);
}

static void cleanup(void)
{
	close(socket_fds[0]);
	close(socket_fds[1]);

	tst_fzsync_pair_cleanup(&fzsync_pair);
}

static void *send_and_close(void *arg)
{
	while (tst_fzsync_run_b(&fzsync_pair)) {

		tst_fzsync_wait_b(&fzsync_pair);
		send(socket_fds[0], MSG, sizeof(MSG), 0);
		send(socket_fds[0], MSG, sizeof(MSG), 0);

		close(socket_fds[0]);

		tst_fzsync_start_race_b(&fzsync_pair);
		close(socket_fds[1]);
		tst_fzsync_end_race_b(&fzsync_pair);
	}
	return arg;
}

static void run(void)
{
	int stat, too_early_count = 0;

	msghdrs[0].msg_hdr.msg_iov->iov_base = (void *)&rbuf;

	tst_fzsync_pair_reset(&fzsync_pair, send_and_close);
	while (tst_fzsync_run_a(&fzsync_pair)) {

		SAFE_SOCKETPAIR(AF_LOCAL, SOCK_DGRAM, 0, (int *)socket_fds);
		tst_fzsync_wait_a(&fzsync_pair);

		tst_fzsync_start_race_a(&fzsync_pair);
		stat = tst_syscall(__NR_recvmmsg,
				   socket_fds[1], msghdrs, 2, 0, &timeout);
		tst_fzsync_end_race_a(&fzsync_pair);

		if (stat == 0)
			tst_res(TWARN, "No messages received, should be one");
		else if (stat < 0) {
			if (errno != EBADF) {
				tst_res(TWARN | TERRNO,
					"recvmmsg failed unexpectedly");
			} else {
				tst_fzsync_pair_add_bias(&fzsync_pair, 1);
				too_early_count++;
			}
		}
	}

	tst_res(TPASS, "Nothing bad happened, probably");
	tst_res(TINFO, "Socket was closed too early %d times", too_early_count);
}

static struct tst_test test = {
	.test_all = run,
	.setup = setup,
	.cleanup = cleanup,
	.max_runtime = 60,
	.tags = (const struct tst_tag[]) {
		{"linux-git", "a2e2725541fa"},
		{"CVE", "2016-7117"},
		{}
	}
};