ossl_asn1_string_to_time_t fails on NonStop in 3.3.0

[rsbeckerca]

I have encountered an issue with ossl_asn1_string_to_time_t having issues in a big-endian x86_64 situation. What I am seeing is below:

90-test_asn1_time.t .....................
        # ERROR: (time_t) 'testdateutc == asn1_to_utc[idx].expected' failed @ /home/ituglib/randall/openssl-3.3.0/test/asn1_time_test.c:439
        # [210327235959Z] compared to [210328005959Z]
        # INFO:  @ /home/ituglib/randall/openssl-3.3.0/test/asn1_time_test.c:443
        # ossl_asn1_string_to_time_t (210328005959Z) failed: expected 1616893199, got 1616889599
        #
        # OPENSSL_TEST_RAND_SEED=1713349126
        not ok 74 - iteration 1
# ------------------------------------------------------------------------------
        # ERROR: (time_t) 'testdateutc == asn1_to_utc[idx].expected' failed @ /home/ituglib/randall/openssl-3.3.0/test/asn1_time_test.c:439
        # [210328000000Z] compared to [210328010000Z]
        # INFO:  @ /home/ituglib/randall/openssl-3.3.0/test/asn1_time_test.c:443
        # ossl_asn1_string_to_time_t (210328010000Z) failed: expected 1616893200, got 1616889600
        #
        # OPENSSL_TEST_RAND_SEED=1713349126
        not ok 75 - iteration 2
# ------------------------------------------------------------------------------
        # ERROR: (time_t) 'testdateutc == asn1_to_utc[idx].expected' failed @ /home/ituglib/randall/openssl-3.3.0/test/asn1_time_test.c:439
        # [210327235959Z] compared to [210328005959Z]
        # INFO:  @ /home/ituglib/randall/openssl-3.3.0/test/asn1_time_test.c:443
        # ossl_asn1_string_to_time_t (20210328015959+0100) failed: expected 1616893199, got 1616889599
        #
        # OPENSSL_TEST_RAND_SEED=1713349126
        not ok 76 - iteration 3
# ------------------------------------------------------------------------------
        # ERROR: (time_t) 'testdateutc == asn1_to_utc[idx].expected' failed @ /home/ituglib/randall/openssl-3.3.0/test/asn1_time_test.c:439
        # [210328000000Z] compared to [210328010000Z]
        # INFO:  @ /home/ituglib/randall/openssl-3.3.0/test/asn1_time_test.c:443
        # ossl_asn1_string_to_time_t (20210328030000+0200) failed: expected 1616893200, got 1616889600
        #
        # OPENSSL_TEST_RAND_SEED=1713349126
        not ok 77 - iteration 4
# ------------------------------------------------------------------------------
    # OPENSSL_TEST_RAND_SEED=1713349126
    not ok 7 - convert_asn1_to_time_t
# ------------------------------------------------------------------------------
../../util/wrap.pl ../../test/asn1_time_test => 255
not ok 1 - running asn1_time_test

[t8m]

AFAIK we do not see the issue on linux big endian cross-compiled targets so this is most probably something NonStop specific.

[rsbeckerca]

AFAIK we do not see the issue on linux big endian cross-compiled targets so this is most probably something NonStop specific.

The difference is exactly 1 hour. This looks like a timezone problem. I am checking whether this is a system setup problem or exists across multiple boxes in different timezones.

[rsbeckerca]

This is not a valid assumption (although unrelated to the case):

#if ((ULONG_MAX >> 31) >> 31) >= 1

time_t can be 32 bits on some platforms. We have a specific need to support 32-bit builds. Has OpenSSL specifically stated no 32-bit time_t support?

[rsbeckerca]

If I set export TZ=MDT the test passes. I suspect the time configuration on the test box I am using may be wrong.

[levitte]

This is not a valid assumption (although unrelated to the case):

#if ((ULONG_MAX >> 31) >> 31) >= 1

time_t can be 32 bits on some platforms. We have a specific need to support 32-bit builds. Has OpenSSL specifically stated no 32-bit time_t support?

That's in this section of code, in a different test than the one failing:

openssl/test/asn1_time_test.c

Lines 449 to 471 in 4e3c1e6

	/*
	* this test is here to exercise ossl_asn1_time_from_tm
	* with an integer year close to INT_MAX.
	*/
	static int convert_tm_to_asn1_time(void)
	{
	/* we need 64 bit time_t */
	#if ((ULONG_MAX >> 31) >> 31) >= 1
	time_t t;
	ASN1_TIME *at;
	if (sizeof(time_t) * CHAR_BIT >= 64) {
	t = 67768011791126057ULL;
	at = ASN1_TIME_set(NULL, t);
	/*
	* If ASN1_TIME_set returns NULL, it means it could not handle the input
	* which is fine for this edge case.
	*/
	ASN1_STRING_free(at);
	}
	#endif
	return 1;
	}

The failing test is this function, which doesn't have the 64-bit requirement as far as I can see:

openssl/test/asn1_time_test.c

Lines 433 to 447 in 4e3c1e6

	static int convert_asn1_to_time_t(int idx)
	{
	time_t testdateutc;
	testdateutc = ossl_asn1_string_to_time_t(asn1_to_utc[idx].input);
	if (!TEST_time_t_eq(testdateutc, asn1_to_utc[idx].expected)) {
	TEST_info("ossl_asn1_string_to_time_t (%s) failed: expected %lli, got %lli\n",
	asn1_to_utc[idx].input,
	(long long int)asn1_to_utc[idx].expected,
	(long long int)testdateutc);
	return 0;
	}
	return 1;
	}

If I set export TZ=MDT the test passes. I suspect the time configuration on the test box I am using may be wrong.

That starts to seem likely. Does the command date (if you have that command) give you a clue?

[rsbeckerca]

I am waiting on the person who configured the system to get back to me with DST table information. The system is only 2 days old, so there are likely issues. Will report back.