tomcrypt/tests/base64_test.c
2017-06-20 15:16:11 +02:00

135 lines
4.3 KiB
C

/* LibTomCrypt, modular cryptographic library -- Tom St Denis
*
* LibTomCrypt is a library that provides various cryptographic
* algorithms in a highly modular and flexible manner.
*
* The library is free for all purposes without any express
* guarantee it works.
*/
#include <tomcrypt_test.h>
#if defined(LTC_BASE64) || defined(LTC_BASE64_URL)
int base64_test(void)
{
unsigned char in[64], out[256], tmp[64];
unsigned long x, l1, l2, slen1;
const unsigned char special_case[] = {
0xbe, 0xe8, 0x92, 0x3c, 0xa2, 0x25, 0xf0, 0xf8,
0x91, 0xe4, 0xef, 0xab, 0x0b, 0x8c, 0xfd, 0xff,
0x14, 0xd0, 0x29, 0x9d, 0x00 };
#if defined(LTC_BASE64)
/*
TEST CASES SOURCE:
Network Working Group S. Josefsson
Request for Comments: 4648 SJD
Obsoletes: 3548 October 2006
Category: Standards Track
*/
const struct {
const char* s;
const char* b64;
} cases[] = {
{"", "" },
{"f", "Zg==" },
{"fo", "Zm8=" },
{"foo", "Zm9v" },
{"foob", "Zm9vYg==" },
{"fooba", "Zm9vYmE=" },
{"foobar", "Zm9vYmFy"},
{(char*)special_case,"vuiSPKIl8PiR5O+rC4z9/xTQKZ0="}
};
#endif
#ifdef LTC_BASE64_URL
const struct {
const char* s;
int is_strict;
} url_cases[] = {
{"vuiSPKIl8PiR5O-rC4z9_xTQKZ0", 0},
{"vuiSPKIl8PiR5O-rC4z9_xTQKZ0=", 1},
{"vuiS*PKIl8P*iR5O-rC4*z9_xTQKZ0", 0},
{"vuiS*PKIl8P*iR5O-rC4*z9_xTQKZ0=", 0},
{"vuiS*PKIl8P*iR5O-rC4*z9_xTQKZ0==", 0},
{"vuiS*PKIl8P*iR5O-rC4*z9_xTQKZ0===", 0},
{"vuiS*PKIl8P*iR5O-rC4*z9_xTQKZ0====", 0},
{"vuiS*=PKIl8P*iR5O-rC4*z9_xTQKZ0=", 0},
{"vuiS*==PKIl8P*iR5O-rC4*z9_xTQKZ0=", 0},
{"vuiS*===PKIl8P*iR5O-rC4*z9_xTQKZ0=", 0},
};
for (x = 0; x < sizeof(url_cases)/sizeof(url_cases[0]); ++x) {
slen1 = strlen(url_cases[x].s);
l1 = sizeof(out);
if(url_cases[x].is_strict)
DO(base64url_strict_decode((unsigned char*)url_cases[x].s, slen1, out, &l1));
else
DO(base64url_decode((unsigned char*)url_cases[x].s, slen1, out, &l1));
if (compare_testvector(out, l1, special_case, sizeof(special_case) - 1, "base64url decode", x)) {
return 1;
}
if(x < 2) {
l2 = sizeof(tmp);
if(x == 0)
DO(base64url_encode(out, l1, tmp, &l2));
else
DO(base64url_strict_encode(out, l1, tmp, &l2));
if (compare_testvector(tmp, l2, url_cases[x].s, strlen(url_cases[x].s), "base64url encode", x)) {
return 1;
}
}
}
DO(base64url_strict_decode((unsigned char*)url_cases[4].s, slen1, out, &l1) == CRYPT_INVALID_PACKET ? CRYPT_OK : CRYPT_INVALID_PACKET);
#endif
#if defined(LTC_BASE64)
for (x = 0; x < sizeof(cases)/sizeof(cases[0]); ++x) {
memset(out, 0, sizeof(out));
memset(tmp, 0, sizeof(tmp));
slen1 = strlen(cases[x].s);
l1 = sizeof(out);
DO(base64_encode((unsigned char*)cases[x].s, slen1, out, &l1));
l2 = sizeof(tmp);
DO(base64_strict_decode(out, l1, tmp, &l2));
if (compare_testvector(out, l1, cases[x].b64, strlen(cases[x].b64), "base64 encode", x) ||
compare_testvector(tmp, l2, cases[x].s, slen1, "base64 decode", x)) {
return 1;
}
}
for (x = 0; x < 64; x++) {
yarrow_read(in, x, &yarrow_prng);
l1 = sizeof(out);
DO(base64_encode(in, x, out, &l1));
l2 = sizeof(tmp);
DO(base64_decode(out, l1, tmp, &l2));
if (compare_testvector(tmp, x, in, x, "random base64", x)) {
return 1;
}
}
x--;
memmove(&out[11], &out[10], l1 - 10);
out[10] = '=';
l1++;
l2 = sizeof(tmp);
DO(base64_decode(out, l1, tmp, &l2));
if (compare_testvector(tmp, l2, in, l2, "relaxed base64 decoding", -1)) {
print_hex("input ", out, l1);
return 1;
}
l2 = sizeof(tmp);
DO(base64_strict_decode(out, l1, tmp, &l2) == CRYPT_INVALID_PACKET ? CRYPT_OK : CRYPT_INVALID_PACKET);
#endif
return 0;
}
#endif
/* ref: $Format:%D$ */
/* git commit: $Format:%H$ */
/* commit time: $Format:%ai$ */