[Bug? - HRT] clock_settime w/tp out of bounds

Fleischer, Julie N (julie.n.fleischer@intel.com)
Wed, 4 Dec 2002 11:24:43 -0800


Here's another bug report for the high-res-timers project from my running
the POSIX Test Suite Timers tests against the high-res-timers
implementation.

Description: clock_settime() is not failing as expected when tp has a
nanosecond value < 0 or >= 1000 million.

Details:
The POSIX spec (2001 - lines 6659-6660; 1993 - lines 127-128) states that if
a struct timespec *tp is passed that has a nanosecond value < 0 or >= 1000
million, the clock_settime() function will fail with errno EINVAL and return
-1.

In the high-res-timers implementation, if tp.tv_nsec is < 0 or >= 1000
million, the function returns 0 and sets the time as if tv_nsec were a valid
value.

Additional information is below:
kernel used = 2.5.50
HRT patches applied =
hrtimers-posix-2.5.50-1.0.patch
hrtimers-core-2.5.50-1.0.patch
hrtimers-i386-2.5.50-1.0.patch
hrtimers-hrposix-2.5.50-1.0.patch
relevant lines of .config =
CONFIG_HIGH_RES_TIMERS=y
# CONFIG_HIGH_RES_TIMER_ACPI_PM is not set
CONFIG_HIGH_RES_TIMER_TSC=y
# CONFIG_HIGH_RES_TIMER_PIT is not set

I have created one large test case which reproduces the issue on various
nsec values < 0 and >= 1000 (19-1.c). Note: Because it actually sets the
time, if it is run, you'll need to sync your time back to the correct time
at the end.

Compiled with:
gcc -g -O2 -Wall -Werror 19-1.c -L/lib/libposixtime.so -lposixtime

**These views are not necessarily those of my employer.**

Test 19-1.c - also on-line at (sorry about the breaks):
http://cvs.sourceforge.net/cgi-bin/viewcvs.cgi/posixtest/posixtestsuite/conf
ormance/behavior/clock_settime/19-1.c?rev=HEAD
-----------

/*
* Copyright (c) 2002, Intel Corporation. All rights reserved.
* Created by: julie.n.fleischer REMOVE-THIS AT intel DOT com
* This file is licensed under the GPL license. For the full content
* of this license, see the COPYING file at the top level of this
* source tree.
*
* Test that clock_settime() sets errno=EINVAL if tp has a nsec value < 0 or

* >= 1000 million.
*
* Test calling clock_settime() with the following tp.tv_nsec values:
* MIN INT = INT32_MIN
* MAX INT = INT32_MAX
* MIN INT - 1 = 2147483647 (this is what gcc will set to)
* MAX INT + 1 = -2147483647 (this is what gcc will set to)
* unassigned value = -1073743192 (ex. of what gcc will set to)
* unassigned value = 1073743192 (ex. of what gcc will set to)
* -1
* 1000000000
* 1000000001
*
* The clock_id CLOCK_REALTIME is used.
*/
#include <stdio.h>
#include <time.h>
#include <errno.h>
#include <stdint.h>

#define PTS_PASS 0
#define PTS_FAIL 1
#define PTS_UNRESOLVED 2

#define NUMINVALIDTESTS 9

static int invalid_tests[NUMINVALIDTESTS] = {
INT32_MIN, INT32_MAX, 2147483647, -2147483647, -1073743192,
1073743192, -1, 1000000000, 1000000001
};

int main(int argc, char *argv[])
{
struct timespec tsset, tscurrent;
int i;
int failure = 0;

if (clock_gettime(CLOCK_REALTIME, &tscurrent) != 0) {
printf("clock_gettime() did not return success\n");
return PTS_UNRESOLVED;
}

for (i = 0; i < NUMINVALIDTESTS; i++) {
tsset.tv_sec = tscurrent.tv_sec;
tsset.tv_nsec = invalid_tests[i];

printf("Test %d sec %d nsec\n",
(int) tsset.tv_sec, (int) tsset.tv_nsec);
if (clock_settime(CLOCK_REALTIME, &tsset) == -1) {
if (EINVAL != errno) {
printf("errno != EINVAL\n");
failure = 1;
}
} else {
printf("clock_settime() did not return -1\n");
failure = 1;
}
}

if (failure) {
printf("At least one test FAILED -- see above\n");
return PTS_FAIL;
} else {
printf("All tests PASSED\n");
return PTS_PASS;
}

return PTS_UNRESOLVED;
}
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/