mbedtls/tests/suites/test_suite_asn1parse.function

796 lines
24 KiB
Text
Raw Normal View History

/* BEGIN_HEADER */
#include <errno.h>
#include <stdlib.h>
#include <limits.h>
#include "mbedtls/bignum.h"
#include "mbedtls/asn1.h"
#if defined(MBEDTLS_ASN1_WRITE_C)
#include "mbedtls/asn1write.h"
#endif
/* Used internally to report an error that indicates a bug in a parsing function. */
#define ERR_PARSE_INCONSISTENCY INT_MAX
/* Use this magic value in some tests to indicate that the expected result
* should not be checked. */
#define UNPREDICTABLE_RESULT 0x5552
static int nested_parse( unsigned char **const p,
const unsigned char *const end )
{
int ret;
size_t len = 0;
size_t len2 = 0;
unsigned char *const start = *p;
unsigned char *content_start;
unsigned char tag;
/* First get the length, skipping over the tag. */
content_start = start + 1;
ret = mbedtls_asn1_get_len( &content_start, end, &len );
TEST_ASSERT( content_start <= end );
if( ret != 0 )
return( ret );
/* Since we have a valid element start (tag and length), retrieve and
* check the tag. */
tag = start[0];
TEST_EQUAL( mbedtls_asn1_get_tag( p, end, &len2, tag ^ 1 ),
MBEDTLS_ERR_ASN1_UNEXPECTED_TAG );
*p = start;
TEST_EQUAL( mbedtls_asn1_get_tag( p, end, &len2, tag ), 0 );
TEST_EQUAL( len, len2 );
TEST_ASSERT( *p == content_start );
*p = content_start;
switch( tag & 0x1f )
{
case MBEDTLS_ASN1_BOOLEAN:
{
int val = -257;
*p = start;
ret = mbedtls_asn1_get_bool( p, end, &val );
if( ret == 0 )
TEST_ASSERT( val == 0 || val == 1 );
break;
}
case MBEDTLS_ASN1_INTEGER:
{
#if defined(MBEDTLS_BIGNUM_C)
mbedtls_mpi mpi;
mbedtls_mpi_init( &mpi );
*p = start;
ret = mbedtls_asn1_get_mpi( p, end, &mpi );
mbedtls_mpi_free( &mpi );
#else
*p = start + 1;
ret = mbedtls_asn1_get_len( p, end, &len );
*p += len;
#endif
/* If we're sure that the number fits in an int, also
* call mbedtls_asn1_get_int(). */
if( ret == 0 && len < sizeof( int ) )
{
int val = -257;
unsigned char *q = start;
ret = mbedtls_asn1_get_int( &q, end, &val );
TEST_ASSERT( *p == q );
}
break;
}
case MBEDTLS_ASN1_BIT_STRING:
{
mbedtls_asn1_bitstring bs;
*p = start;
ret = mbedtls_asn1_get_bitstring( p, end, &bs );
break;
}
case MBEDTLS_ASN1_SEQUENCE:
{
while( *p <= end && *p < content_start + len && ret == 0 )
ret = nested_parse( p, content_start + len );
break;
}
case MBEDTLS_ASN1_OCTET_STRING:
case MBEDTLS_ASN1_NULL:
case MBEDTLS_ASN1_OID:
case MBEDTLS_ASN1_UTF8_STRING:
case MBEDTLS_ASN1_SET:
case MBEDTLS_ASN1_PRINTABLE_STRING:
case MBEDTLS_ASN1_T61_STRING:
case MBEDTLS_ASN1_IA5_STRING:
case MBEDTLS_ASN1_UTC_TIME:
case MBEDTLS_ASN1_GENERALIZED_TIME:
case MBEDTLS_ASN1_UNIVERSAL_STRING:
case MBEDTLS_ASN1_BMP_STRING:
default:
/* No further testing implemented for this tag. */
*p += len;
return( 0 );
}
TEST_ASSERT( *p <= end );
return( ret );
exit:
return( ERR_PARSE_INCONSISTENCY );
}
int get_len_step( const data_t *input, size_t buffer_size,
size_t actual_length )
{
unsigned char *buf = NULL;
unsigned char *p = NULL;
unsigned char *end;
size_t parsed_length;
int ret;
mbedtls_test_set_step( buffer_size );
/* Allocate a new buffer of exactly the length to parse each time.
* This gives memory sanitizers a chance to catch buffer overreads. */
if( buffer_size == 0 )
{
ASSERT_ALLOC( buf, 1 );
end = buf + 1;
p = end;
}
else
{
ASSERT_ALLOC_WEAK( buf, buffer_size );
if( buffer_size > input->len )
{
memcpy( buf, input->x, input->len );
memset( buf + input->len, 'A', buffer_size - input->len );
}
else
{
memcpy( buf, input->x, buffer_size );
}
p = buf;
end = buf + buffer_size;
}
ret = mbedtls_asn1_get_len( &p, end, &parsed_length );
if( buffer_size >= input->len + actual_length )
{
TEST_EQUAL( ret, 0 );
TEST_ASSERT( p == buf + input->len );
TEST_EQUAL( parsed_length, actual_length );
}
else
{
TEST_EQUAL( ret, MBEDTLS_ERR_ASN1_OUT_OF_DATA );
}
mbedtls_free( buf );
return( 1 );
exit:
mbedtls_free( buf );
return( 0 );
}
typedef struct
{
const unsigned char *input_start;
const char *description;
} traverse_state_t;
/* Value returned by traverse_callback if description runs out. */
#define RET_TRAVERSE_STOP 1
/* Value returned by traverse_callback if description has an invalid format
* (see traverse_sequence_of). */
#define RET_TRAVERSE_ERROR 2
static int traverse_callback( void *ctx, int tag,
unsigned char *content, size_t len )
{
traverse_state_t *state = ctx;
size_t offset;
const char *rest = state->description;
unsigned long n;
TEST_ASSERT( content > state->input_start );
offset = content - state->input_start;
mbedtls_test_set_step( offset );
if( *rest == 0 )
return( RET_TRAVERSE_STOP );
n = strtoul( rest, (char **) &rest, 0 );
TEST_EQUAL( n, offset );
TEST_EQUAL( *rest, ',' );
++rest;
n = strtoul( rest, (char **) &rest, 0 );
TEST_EQUAL( n, (unsigned) tag );
TEST_EQUAL( *rest, ',' );
++rest;
n = strtoul( rest, (char **) &rest, 0 );
TEST_EQUAL( n, len );
if( *rest == ',' )
++rest;
state->description = rest;
return( 0 );
exit:
return( RET_TRAVERSE_ERROR );
}
/* END_HEADER */
/* BEGIN_DEPENDENCIES
* depends_on:MBEDTLS_ASN1_PARSE_C
* END_DEPENDENCIES
*/
/* BEGIN_CASE */
void parse_prefixes( const data_t *input,
int full_result,
int overfull_result )
{
/* full_result: expected result from parsing the given string. */
/* overfull_result: expected_result from parsing the given string plus
* some trailing garbage. This may be UNPREDICTABLE_RESULT to accept
* any result: use this for invalid inputs that may or may not become
* valid depending on what the trailing garbage is. */
unsigned char *buf = NULL;
unsigned char *p = NULL;
size_t buffer_size;
int ret;
/* Test every prefix of the input, except the empty string.
* The first byte of the string is the tag. Without a tag byte,
* we wouldn't know what to parse the input as.
* Also test the input followed by an extra byte.
*/
for( buffer_size = 1; buffer_size <= input->len + 1; buffer_size++ )
{
mbedtls_test_set_step( buffer_size );
/* Allocate a new buffer of exactly the length to parse each time.
* This gives memory sanitizers a chance to catch buffer overreads. */
ASSERT_ALLOC( buf, buffer_size );
memcpy( buf, input->x, buffer_size );
p = buf;
ret = nested_parse( &p, buf + buffer_size );
if( ret == ERR_PARSE_INCONSISTENCY )
goto exit;
if( buffer_size < input->len )
{
TEST_EQUAL( ret, MBEDTLS_ERR_ASN1_OUT_OF_DATA );
}
else if( buffer_size == input->len )
{
TEST_EQUAL( ret, full_result );
}
else /* ( buffer_size > input->len ) */
{
if( overfull_result != UNPREDICTABLE_RESULT )
TEST_EQUAL( ret, overfull_result );
}
if( ret == 0 )
TEST_ASSERT( p == buf + input->len );
mbedtls_free( buf );
buf = NULL;
}
exit:
mbedtls_free( buf );
}
/* END_CASE */
/* BEGIN_CASE */
void get_len( const data_t *input, int actual_length_arg )
{
size_t actual_length = actual_length_arg;
size_t buffer_size;
/* Test prefixes of a buffer containing the given length string
* followed by `actual_length` bytes of payload. To save a bit of
* time, we skip some "boring" prefixes: we don't test prefixes where
* the payload is truncated more than one byte away from either end,
* and we only test the empty string on a 1-byte input.
*/
for( buffer_size = 1; buffer_size <= input->len + 1; buffer_size++ )
{
if( ! get_len_step( input, buffer_size, actual_length ) )
goto exit;
}
if( ! get_len_step( input, input->len + actual_length - 1, actual_length ) )
goto exit;
if( ! get_len_step( input, input->len + actual_length, actual_length ) )
goto exit;
}
/* END_CASE */
/* BEGIN_CASE */
void get_boolean( const data_t *input,
int expected_value, int expected_result )
{
unsigned char *p = input->x;
int val;
int ret;
ret = mbedtls_asn1_get_bool( &p, input->x + input->len, &val );
TEST_EQUAL( ret, expected_result );
if( expected_result == 0 )
{
TEST_EQUAL( val, expected_value );
TEST_ASSERT( p == input->x + input->len );
}
}
/* END_CASE */
/* BEGIN_CASE */
void empty_integer( const data_t *input )
{
unsigned char *p;
#if defined(MBEDTLS_BIGNUM_C)
mbedtls_mpi actual_mpi;
#endif
int val;
#if defined(MBEDTLS_BIGNUM_C)
mbedtls_mpi_init( & actual_mpi );
#endif
/* An INTEGER with no content is not valid. */
p = input->x;
TEST_EQUAL( mbedtls_asn1_get_int( &p, input->x + input->len, &val ),
MBEDTLS_ERR_ASN1_INVALID_LENGTH );
#if defined(MBEDTLS_BIGNUM_C)
/* INTEGERs are sometimes abused as bitstrings, so the library accepts
* an INTEGER with empty content and gives it the value 0. */
p = input->x;
TEST_EQUAL( mbedtls_asn1_get_mpi( &p, input->x + input->len, &actual_mpi ),
0 );
TEST_EQUAL( mbedtls_mpi_cmp_int( &actual_mpi, 0 ), 0 );
#endif
exit:
#if defined(MBEDTLS_BIGNUM_C)
mbedtls_mpi_free( &actual_mpi );
#endif
/*empty cleanup in some configurations*/ ;
}
/* END_CASE */
/* BEGIN_CASE */
void get_integer( const data_t *input,
const char *expected_hex, int expected_result )
{
unsigned char *p;
#if defined(MBEDTLS_BIGNUM_C)
mbedtls_mpi expected_mpi;
mbedtls_mpi actual_mpi;
mbedtls_mpi complement;
int expected_result_for_mpi = expected_result;
#endif
long expected_value;
int expected_result_for_int = expected_result;
int val;
int ret;
#if defined(MBEDTLS_BIGNUM_C)
mbedtls_mpi_init( &expected_mpi );
mbedtls_mpi_init( &actual_mpi );
mbedtls_mpi_init( &complement );
#endif
errno = 0;
expected_value = strtol( expected_hex, NULL, 16 );
if( expected_result == 0 &&
( errno == ERANGE
#if LONG_MAX > INT_MAX
|| expected_value > INT_MAX || expected_value < INT_MIN
#endif
) )
{
/* The library returns the dubious error code INVALID_LENGTH
* for integers that are out of range. */
expected_result_for_int = MBEDTLS_ERR_ASN1_INVALID_LENGTH;
}
if( expected_result == 0 && expected_value < 0 )
{
/* The library does not support negative INTEGERs and
* returns the dubious error code INVALID_LENGTH.
* Test that we preserve the historical behavior. If we
* decide to change the behavior, we'll also change this test. */
expected_result_for_int = MBEDTLS_ERR_ASN1_INVALID_LENGTH;
}
p = input->x;
ret = mbedtls_asn1_get_int( &p, input->x + input->len, &val );
TEST_EQUAL( ret, expected_result_for_int );
if( ret == 0 )
{
TEST_EQUAL( val, expected_value );
TEST_ASSERT( p == input->x + input->len );
}
#if defined(MBEDTLS_BIGNUM_C)
ret = mbedtls_test_read_mpi( &expected_mpi, expected_hex );
TEST_ASSERT( ret == 0 || ret == MBEDTLS_ERR_MPI_BAD_INPUT_DATA );
if( ret == MBEDTLS_ERR_MPI_BAD_INPUT_DATA )
{
/* The data overflows the maximum MPI size. */
expected_result_for_mpi = MBEDTLS_ERR_MPI_BAD_INPUT_DATA;
}
p = input->x;
ret = mbedtls_asn1_get_mpi( &p, input->x + input->len, &actual_mpi );
TEST_EQUAL( ret, expected_result_for_mpi );
if( ret == 0 )
{
if( expected_value >= 0 )
{
TEST_ASSERT( mbedtls_mpi_cmp_mpi( &actual_mpi,
&expected_mpi ) == 0 );
}
else
{
/* The library ignores the sign bit in ASN.1 INTEGERs
* (which makes sense insofar as INTEGERs are sometimes
* abused as bit strings), so the result of parsing them
* is a positive integer such that expected_mpi +
* actual_mpi = 2^n where n is the length of the content
* of the INTEGER. (Leading ff octets don't matter for the
* expected value, but they matter for the actual value.)
* Test that we don't change from this behavior. If we
* decide to fix the library to change the behavior on
* negative INTEGERs, we'll fix this test code. */
unsigned char *q = input->x + 1;
size_t len;
TEST_ASSERT( mbedtls_asn1_get_len( &q, input->x + input->len,
&len ) == 0 );
TEST_ASSERT( mbedtls_mpi_lset( &complement, 1 ) == 0 );
TEST_ASSERT( mbedtls_mpi_shift_l( &complement, len * 8 ) == 0 );
TEST_ASSERT( mbedtls_mpi_add_mpi( &complement, &complement,
&expected_mpi ) == 0 );
TEST_ASSERT( mbedtls_mpi_cmp_mpi( &complement,
&actual_mpi ) == 0 );
}
TEST_ASSERT( p == input->x + input->len );
}
#endif
exit:
#if defined(MBEDTLS_BIGNUM_C)
mbedtls_mpi_free( &expected_mpi );
mbedtls_mpi_free( &actual_mpi );
mbedtls_mpi_free( &complement );
#endif
/*empty cleanup in some configurations*/ ;
}
/* END_CASE */
/* BEGIN_CASE */
void get_enum( const data_t *input,
const char *expected_hex, int expected_result )
{
unsigned char *p;
long expected_value;
int expected_result_for_enum = expected_result;
int val;
int ret;
errno = 0;
expected_value = strtol( expected_hex, NULL, 16 );
if( expected_result == 0 &&
( errno == ERANGE
#if LONG_MAX > INT_MAX
|| expected_value > INT_MAX || expected_value < INT_MIN
#endif
) )
{
/* The library returns the dubious error code INVALID_LENGTH
* for integers that are out of range. */
expected_result_for_enum = MBEDTLS_ERR_ASN1_INVALID_LENGTH;
}
if( expected_result == 0 && expected_value < 0 )
{
/* The library does not support negative INTEGERs and
* returns the dubious error code INVALID_LENGTH.
* Test that we preserve the historical behavior. If we
* decide to change the behavior, we'll also change this test. */
expected_result_for_enum = MBEDTLS_ERR_ASN1_INVALID_LENGTH;
}
p = input->x;
ret = mbedtls_asn1_get_enum( &p, input->x + input->len, &val );
TEST_EQUAL( ret, expected_result_for_enum );
if( ret == 0 )
{
TEST_EQUAL( val, expected_value );
TEST_ASSERT( p == input->x + input->len );
}
}
/* END_CASE */
/* BEGIN_CASE depends_on:MBEDTLS_BIGNUM_C */
void get_mpi_too_large( )
{
unsigned char *buf = NULL;
unsigned char *p;
mbedtls_mpi actual_mpi;
size_t too_many_octets =
MBEDTLS_MPI_MAX_LIMBS * sizeof(mbedtls_mpi_uint) + 1;
size_t size = too_many_octets + 6;
mbedtls_mpi_init( &actual_mpi );
ASSERT_ALLOC( buf, size );
buf[0] = 0x02; /* tag: INTEGER */
buf[1] = 0x84; /* 4-octet length */
buf[2] = ( too_many_octets >> 24 ) & 0xff;
buf[3] = ( too_many_octets >> 16 ) & 0xff;
buf[4] = ( too_many_octets >> 8 ) & 0xff;
buf[5] = too_many_octets & 0xff;
buf[6] = 0x01; /* most significant octet */
p = buf;
TEST_EQUAL( mbedtls_asn1_get_mpi( &p, buf + size, &actual_mpi ),
MBEDTLS_ERR_MPI_ALLOC_FAILED );
exit:
mbedtls_mpi_free( &actual_mpi );
mbedtls_free( buf );
}
/* END_CASE */
/* BEGIN_CASE */
void get_bitstring( const data_t *input,
int expected_length, int expected_unused_bits,
int expected_result, int expected_result_null )
{
mbedtls_asn1_bitstring bs = { 0xdead, 0x21, NULL };
unsigned char *p = input->x;
TEST_EQUAL( mbedtls_asn1_get_bitstring( &p, input->x + input->len, &bs ),
expected_result );
if( expected_result == 0 )
{
TEST_EQUAL( bs.len, (size_t) expected_length );
TEST_EQUAL( bs.unused_bits, expected_unused_bits );
TEST_ASSERT( bs.p != NULL );
TEST_EQUAL( bs.p - input->x + bs.len, input->len );
TEST_ASSERT( p == input->x + input->len );
}
p = input->x;
TEST_EQUAL( mbedtls_asn1_get_bitstring_null( &p, input->x + input->len,
&bs.len ),
expected_result_null );
if( expected_result_null == 0 )
{
TEST_EQUAL( bs.len, (size_t) expected_length );
if( expected_result == 0 )
TEST_ASSERT( p == input->x + input->len - bs.len );
}
}
/* END_CASE */
/* BEGIN_CASE */
void get_sequence_of( const data_t *input, int tag,
const char *description,
int expected_result )
{
/* The description string is a comma-separated list of integers.
* For each element in the SEQUENCE in input, description contains
* two integers: the offset of the element (offset from the start
* of input to the tag of the element) and the length of the
* element's contents.
* "offset1,length1,..." */
mbedtls_asn1_sequence head = { { 0, 0, NULL }, NULL };
mbedtls_asn1_sequence *cur;
unsigned char *p = input->x;
const char *rest = description;
unsigned long n;
unsigned int step = 0;
TEST_EQUAL( mbedtls_asn1_get_sequence_of( &p, input->x + input->len,
&head, tag ),
expected_result );
if( expected_result == 0 )
{
TEST_ASSERT( p == input->x + input->len );
if( ! *rest )
{
TEST_EQUAL( head.buf.tag, 0 );
TEST_ASSERT( head.buf.p == NULL );
TEST_EQUAL( head.buf.len, 0 );
TEST_ASSERT( head.next == NULL );
}
else
{
cur = &head;
while( *rest )
{
mbedtls_test_set_step( step );
TEST_ASSERT( cur != NULL );
TEST_EQUAL( cur->buf.tag, tag );
n = strtoul( rest, (char **) &rest, 0 );
TEST_EQUAL( n, (size_t)( cur->buf.p - input->x ) );
++rest;
n = strtoul( rest, (char **) &rest, 0 );
TEST_EQUAL( n, cur->buf.len );
if( *rest )
++rest;
cur = cur->next;
++step;
}
TEST_ASSERT( cur == NULL );
}
}
exit:
mbedtls_asn1_sequence_free( head.next );
}
/* END_CASE */
/* BEGIN_CASE */
void traverse_sequence_of( const data_t *input,
int tag_must_mask, int tag_must_val,
int tag_may_mask, int tag_may_val,
const char *description,
int expected_result )
{
/* The description string is a comma-separated list of integers.
* For each element in the SEQUENCE in input, description contains
* three integers: the offset of the element's content (offset from
* the start of input to the content of the element), the element's tag,
* and the length of the element's contents.
* "offset1,tag1,length1,..." */
unsigned char *p = input->x;
traverse_state_t traverse_state = {input->x, description};
int ret;
ret = mbedtls_asn1_traverse_sequence_of( &p, input->x + input->len,
(uint8_t) tag_must_mask, (uint8_t) tag_must_val,
(uint8_t) tag_may_mask, (uint8_t) tag_may_val,
traverse_callback, &traverse_state );
if( ret == RET_TRAVERSE_ERROR )
goto exit;
TEST_EQUAL( ret, expected_result );
TEST_EQUAL( *traverse_state.description, 0 );
}
/* END_CASE */
/* BEGIN_CASE */
void get_alg( const data_t *input,
int oid_offset, int oid_length,
int params_tag, int params_offset, int params_length,
int total_length,
int expected_result )
{
mbedtls_asn1_buf oid = { -1, 0, NULL };
mbedtls_asn1_buf params = { -1, 0, NULL };
unsigned char *p = input->x;
int ret;
TEST_EQUAL( mbedtls_asn1_get_alg( &p, input->x + input->len,
&oid, &params ),
expected_result );
if( expected_result == 0 )
{
TEST_EQUAL( oid.tag, MBEDTLS_ASN1_OID );
TEST_EQUAL( oid.p - input->x, oid_offset );
TEST_EQUAL( oid.len, (size_t) oid_length );
TEST_EQUAL( params.tag, params_tag );
if( params_offset != 0 )
TEST_EQUAL( params.p - input->x, params_offset );
else
TEST_ASSERT( params.p == NULL );
TEST_EQUAL( params.len, (size_t) params_length );
TEST_EQUAL( p - input->x, total_length );
}
ret = mbedtls_asn1_get_alg_null( &p, input->x + input->len, &oid );
if( expected_result == 0 && params_offset == 0 )
{
TEST_EQUAL( oid.tag, MBEDTLS_ASN1_OID );
TEST_EQUAL( oid.p - input->x, oid_offset );
TEST_EQUAL( oid.len, (size_t) oid_length );
TEST_EQUAL( p - input->x, total_length );
}
else
TEST_ASSERT( ret != 0 );
}
/* END_CASE */
/* BEGIN_CASE */
void find_named_data( data_t *oid0, data_t *oid1, data_t *oid2, data_t *oid3,
data_t *needle, int from, int position )
{
mbedtls_asn1_named_data nd[] ={
{ {0x06, oid0->len, oid0->x}, {0, 0, NULL}, NULL, 0 },
{ {0x06, oid1->len, oid1->x}, {0, 0, NULL}, NULL, 0 },
{ {0x06, oid2->len, oid2->x}, {0, 0, NULL}, NULL, 0 },
{ {0x06, oid3->len, oid3->x}, {0, 0, NULL}, NULL, 0 },
};
mbedtls_asn1_named_data *pointers[ARRAY_LENGTH( nd ) + 1];
size_t i;
const mbedtls_asn1_named_data *found;
for( i = 0; i < ARRAY_LENGTH( nd ); i++ )
pointers[i] = &nd[i];
pointers[ARRAY_LENGTH( nd )] = NULL;
for( i = 0; i < ARRAY_LENGTH( nd ); i++ )
nd[i].next = pointers[i+1];
found = mbedtls_asn1_find_named_data( (const mbedtls_asn1_named_data*) pointers[from],
(const char *) needle->x,
needle->len );
TEST_ASSERT( found == pointers[position] );
}
/* END_CASE */
/* BEGIN_CASE */
void free_named_data_null( )
{
mbedtls_asn1_free_named_data( NULL );
goto exit; /* Silence unused label warning */
}
/* END_CASE */
/* BEGIN_CASE */
void free_named_data( int with_oid, int with_val, int with_next )
{
mbedtls_asn1_named_data next =
{ {0x06, 0, NULL}, {0, 0xcafe, NULL}, NULL, 0 };
mbedtls_asn1_named_data head =
{ {0x06, 0, NULL}, {0, 0, NULL}, NULL, 0 };
if( with_oid )
ASSERT_ALLOC( head.oid.p, 1 );
if( with_val )
ASSERT_ALLOC( head.val.p, 1 );
if( with_next )
head.next = &next;
mbedtls_asn1_free_named_data( &head );
TEST_ASSERT( head.oid.p == NULL );
TEST_ASSERT( head.val.p == NULL );
TEST_ASSERT( head.next == NULL );
TEST_ASSERT( next.val.len == 0xcafe );
exit:
mbedtls_free( head.oid.p );
mbedtls_free( head.val.p );
}
/* END_CASE */
/* BEGIN_CASE */
void free_named_data_list( int length )
{
mbedtls_asn1_named_data *head = NULL;
int i;
for( i = 0; i < length; i++ )
{
mbedtls_asn1_named_data *new = NULL;
ASSERT_ALLOC( new, sizeof( mbedtls_asn1_named_data ) );
new->next = head;
head = new;
}
mbedtls_asn1_free_named_data_list( &head );
TEST_ASSERT( head == NULL );
/* Most of the point of the test is that it doesn't leak memory.
* So this test is only really useful under a memory leak detection
* framework. */
exit:
mbedtls_asn1_free_named_data_list( &head );
}
/* END_CASE */