mirror of
https://github.com/hashcat/hashcat.git
synced 2024-11-16 04:49:24 +00:00
1372 lines
34 KiB
C
1372 lines
34 KiB
C
/**
|
|
* Author......: See docs/credits.txt
|
|
* License.....: MIT
|
|
*/
|
|
|
|
#include "common.h"
|
|
#include "types.h"
|
|
#include "convert.h"
|
|
#include "shared.h"
|
|
#include "memory.h"
|
|
#include "ext_lzma.h"
|
|
#include <errno.h>
|
|
|
|
#if defined (__CYGWIN__)
|
|
#include <sys/cygwin.h>
|
|
#endif
|
|
|
|
static const char *const PA_000 = "OK";
|
|
static const char *const PA_001 = "Ignored due to comment";
|
|
static const char *const PA_002 = "Ignored due to zero length";
|
|
static const char *const PA_003 = "Line-length exception";
|
|
static const char *const PA_004 = "Hash-length exception";
|
|
static const char *const PA_005 = "Hash-value exception";
|
|
static const char *const PA_006 = "Salt-length exception";
|
|
static const char *const PA_007 = "Salt-value exception";
|
|
static const char *const PA_008 = "Salt-iteration count exception";
|
|
static const char *const PA_009 = "Separator unmatched";
|
|
static const char *const PA_010 = "Signature unmatched";
|
|
static const char *const PA_011 = "Invalid hccapx file size";
|
|
static const char *const PA_012 = "Invalid hccapx eapol size";
|
|
static const char *const PA_013 = "Invalid psafe2 filesize";
|
|
static const char *const PA_014 = "Invalid psafe3 filesize";
|
|
static const char *const PA_015 = "Invalid truecrypt filesize";
|
|
static const char *const PA_016 = "Invalid veracrypt filesize";
|
|
static const char *const PA_017 = "Invalid SIP directive, only MD5 is supported";
|
|
static const char *const PA_018 = "Hash-file exception";
|
|
static const char *const PA_019 = "Hash-encoding exception";
|
|
static const char *const PA_020 = "Salt-encoding exception";
|
|
static const char *const PA_021 = "Invalid LUKS filesize";
|
|
static const char *const PA_022 = "Invalid LUKS identifier";
|
|
static const char *const PA_023 = "Invalid LUKS version";
|
|
static const char *const PA_024 = "Invalid or unsupported LUKS cipher type";
|
|
static const char *const PA_025 = "Invalid or unsupported LUKS cipher mode";
|
|
static const char *const PA_026 = "Invalid or unsupported LUKS hash type";
|
|
static const char *const PA_027 = "Invalid LUKS key size";
|
|
static const char *const PA_028 = "Disabled LUKS key detected";
|
|
static const char *const PA_029 = "Invalid LUKS key AF stripes count";
|
|
static const char *const PA_030 = "Invalid combination of LUKS hash type and cipher type";
|
|
static const char *const PA_031 = "Invalid hccapx signature";
|
|
static const char *const PA_032 = "Invalid hccapx version";
|
|
static const char *const PA_033 = "Invalid hccapx message pair";
|
|
static const char *const PA_034 = "Token encoding exception";
|
|
static const char *const PA_035 = "Token length exception";
|
|
static const char *const PA_036 = "Insufficient entropy exception";
|
|
static const char *const PA_037 = "Hash contains unsupported compression type for current mode";
|
|
static const char *const PA_038 = "Invalid key size";
|
|
static const char *const PA_039 = "Invalid block size";
|
|
static const char *const PA_040 = "Invalid or unsupported cipher";
|
|
static const char *const PA_041 = "Invalid filesize";
|
|
static const char *const PA_042 = "IV length exception";
|
|
static const char *const PA_043 = "CT length exception";
|
|
static const char *const PA_255 = "Unknown error";
|
|
|
|
static const char *const OPTI_STR_OPTIMIZED_KERNEL = "Optimized-Kernel";
|
|
static const char *const OPTI_STR_ZERO_BYTE = "Zero-Byte";
|
|
static const char *const OPTI_STR_PRECOMPUTE_INIT = "Precompute-Init";
|
|
static const char *const OPTI_STR_MEET_IN_MIDDLE = "Meet-In-The-Middle";
|
|
static const char *const OPTI_STR_EARLY_SKIP = "Early-Skip";
|
|
static const char *const OPTI_STR_NOT_SALTED = "Not-Salted";
|
|
static const char *const OPTI_STR_NOT_ITERATED = "Not-Iterated";
|
|
static const char *const OPTI_STR_PREPENDED_SALT = "Prepended-Salt";
|
|
static const char *const OPTI_STR_APPENDED_SALT = "Appended-Salt";
|
|
static const char *const OPTI_STR_SINGLE_HASH = "Single-Hash";
|
|
static const char *const OPTI_STR_SINGLE_SALT = "Single-Salt";
|
|
static const char *const OPTI_STR_BRUTE_FORCE = "Brute-Force";
|
|
static const char *const OPTI_STR_RAW_HASH = "Raw-Hash";
|
|
static const char *const OPTI_STR_SLOW_HASH_SIMD_INIT = "Slow-Hash-SIMD-INIT";
|
|
static const char *const OPTI_STR_SLOW_HASH_SIMD_LOOP = "Slow-Hash-SIMD-LOOP";
|
|
static const char *const OPTI_STR_SLOW_HASH_SIMD_COMP = "Slow-Hash-SIMD-COMP";
|
|
static const char *const OPTI_STR_USES_BITS_8 = "Uses-8-Bit";
|
|
static const char *const OPTI_STR_USES_BITS_16 = "Uses-16-Bit";
|
|
static const char *const OPTI_STR_USES_BITS_32 = "Uses-32-Bit";
|
|
static const char *const OPTI_STR_USES_BITS_64 = "Uses-64-Bit";
|
|
|
|
static const char *const HASH_CATEGORY_UNDEFINED_STR = "Undefined";
|
|
static const char *const HASH_CATEGORY_RAW_HASH_STR = "Raw Hash";
|
|
static const char *const HASH_CATEGORY_RAW_HASH_SALTED_STR = "Raw Hash salted and/or iterated";
|
|
static const char *const HASH_CATEGORY_RAW_HASH_AUTHENTICATED_STR = "Raw Hash authenticated";
|
|
static const char *const HASH_CATEGORY_RAW_CIPHER_KPA_STR = "Raw Cipher, Known-plaintext attack";
|
|
static const char *const HASH_CATEGORY_GENERIC_KDF_STR = "Generic KDF";
|
|
static const char *const HASH_CATEGORY_NETWORK_PROTOCOL_STR = "Network Protocol";
|
|
static const char *const HASH_CATEGORY_FORUM_SOFTWARE_STR = "Forums, CMS, E-Commerce";
|
|
static const char *const HASH_CATEGORY_DATABASE_SERVER_STR = "Database Server";
|
|
static const char *const HASH_CATEGORY_NETWORK_SERVER_STR = "FTP, HTTP, SMTP, LDAP Server";
|
|
static const char *const HASH_CATEGORY_RAW_CHECKSUM_STR = "Raw Checksum";
|
|
static const char *const HASH_CATEGORY_OS_STR = "Operating System";
|
|
static const char *const HASH_CATEGORY_EAS_STR = "Enterprise Application Software (EAS)";
|
|
static const char *const HASH_CATEGORY_ARCHIVE_STR = "Archive";
|
|
static const char *const HASH_CATEGORY_FDE_STR = "Full-Disk Encryption (FDE)";
|
|
static const char *const HASH_CATEGORY_FBE_STR = "File-Based Encryption (FBE)";
|
|
static const char *const HASH_CATEGORY_DOCUMENTS_STR = "Document";
|
|
static const char *const HASH_CATEGORY_PASSWORD_MANAGER_STR = "Password Manager";
|
|
static const char *const HASH_CATEGORY_OTP_STR = "One-Time Password";
|
|
static const char *const HASH_CATEGORY_PLAIN_STR = "Plaintext";
|
|
static const char *const HASH_CATEGORY_FRAMEWORK_STR = "Framework";
|
|
static const char *const HASH_CATEGORY_PRIVATE_KEY_STR = "Private Key";
|
|
static const char *const HASH_CATEGORY_IMS_STR = "Instant Messaging Service";
|
|
static const char *const HASH_CATEGORY_CRYPTOCURRENCY_WALLET_STR = "Cryptocurrency Wallet";
|
|
|
|
int sort_by_string_sized (const void *p1, const void *p2)
|
|
{
|
|
string_sized_t *s1 = (string_sized_t *) p1;
|
|
string_sized_t *s2 = (string_sized_t *) p2;
|
|
|
|
const int d = s1->len - s2->len;
|
|
|
|
if (d != 0) return d;
|
|
|
|
return memcmp (s1->buf, s2->buf, s1->len);
|
|
}
|
|
|
|
int sort_by_stringptr (const void *p1, const void *p2)
|
|
{
|
|
const char* const *s1 = (const char* const *) p1;
|
|
const char* const *s2 = (const char* const *) p2;
|
|
|
|
return strcmp (*s1, *s2);
|
|
}
|
|
|
|
static inline int get_msb32 (const u32 v)
|
|
{
|
|
int i;
|
|
|
|
for (i = 32; i > 0; i--) if ((v >> (i - 1)) & 1) break;
|
|
|
|
return i;
|
|
}
|
|
|
|
static inline int get_msb64 (const u64 v)
|
|
{
|
|
int i;
|
|
|
|
for (i = 64; i > 0; i--) if ((v >> (i - 1)) & 1) break;
|
|
|
|
return i;
|
|
}
|
|
|
|
bool overflow_check_u32_add (const u32 a, const u32 b)
|
|
{
|
|
const int a_msb = get_msb32 (a);
|
|
const int b_msb = get_msb32 (b);
|
|
|
|
return ((a_msb < 32) && (b_msb < 32));
|
|
}
|
|
|
|
bool overflow_check_u32_mul (const u32 a, const u32 b)
|
|
{
|
|
const int a_msb = get_msb32 (a);
|
|
const int b_msb = get_msb32 (b);
|
|
|
|
return ((a_msb + b_msb) < 32);
|
|
}
|
|
|
|
bool overflow_check_u64_add (const u64 a, const u64 b)
|
|
{
|
|
const int a_msb = get_msb64 (a);
|
|
const int b_msb = get_msb64 (b);
|
|
|
|
return ((a_msb < 64) && (b_msb < 64));
|
|
}
|
|
|
|
bool overflow_check_u64_mul (const u64 a, const u64 b)
|
|
{
|
|
const int a_msb = get_msb64 (a);
|
|
const int b_msb = get_msb64 (b);
|
|
|
|
return ((a_msb + b_msb) < 64);
|
|
}
|
|
|
|
bool is_power_of_2 (const u32 v)
|
|
{
|
|
return (v && !(v & (v - 1)));
|
|
}
|
|
|
|
u32 mydivc32 (const u32 dividend, const u32 divisor)
|
|
{
|
|
u32 quotient = dividend / divisor;
|
|
|
|
if (dividend % divisor) quotient++;
|
|
|
|
return quotient;
|
|
}
|
|
|
|
u64 mydivc64 (const u64 dividend, const u64 divisor)
|
|
{
|
|
u64 quotient = dividend / divisor;
|
|
|
|
if (dividend % divisor) quotient++;
|
|
|
|
return quotient;
|
|
}
|
|
|
|
char *filename_from_filepath (char *filepath)
|
|
{
|
|
char *ptr = NULL;
|
|
|
|
if ((ptr = strrchr (filepath, '/')) != NULL)
|
|
{
|
|
ptr++;
|
|
}
|
|
else if ((ptr = strrchr (filepath, '\\')) != NULL)
|
|
{
|
|
ptr++;
|
|
}
|
|
else
|
|
{
|
|
ptr = filepath;
|
|
}
|
|
|
|
return ptr;
|
|
}
|
|
|
|
void naive_replace (char *s, const char key_char, const char replace_char)
|
|
{
|
|
const size_t len = strlen (s);
|
|
|
|
for (size_t in = 0; in < len; in++)
|
|
{
|
|
const char c = s[in];
|
|
|
|
if (c == key_char)
|
|
{
|
|
s[in] = replace_char;
|
|
}
|
|
}
|
|
}
|
|
|
|
void naive_escape (char *s, size_t s_max, const char key_char, const char escape_char)
|
|
{
|
|
char s_escaped[1024] = { 0 };
|
|
|
|
size_t s_escaped_max = sizeof (s_escaped);
|
|
|
|
const size_t len = strlen (s);
|
|
|
|
for (size_t in = 0, out = 0; in < len; in++, out++)
|
|
{
|
|
const char c = s[in];
|
|
|
|
if (c == key_char)
|
|
{
|
|
s_escaped[out] = escape_char;
|
|
|
|
out++;
|
|
}
|
|
|
|
if (out == s_escaped_max - 2) break;
|
|
|
|
s_escaped[out] = c;
|
|
}
|
|
|
|
strncpy (s, s_escaped, s_max - 1);
|
|
}
|
|
|
|
int hc_asprintf (char **strp, const char *fmt, ...)
|
|
{
|
|
va_list args;
|
|
va_start (args, fmt);
|
|
int rc = vasprintf (strp, fmt, args);
|
|
va_end (args);
|
|
return rc;
|
|
}
|
|
|
|
#if defined (_WIN)
|
|
#define __WINDOWS__
|
|
#endif
|
|
#include "sort_r.h"
|
|
#if defined (_WIN)
|
|
#undef __WINDOWS__
|
|
#endif
|
|
|
|
void hc_qsort_r (void *base, size_t nmemb, size_t size, int (*compar) (const void *, const void *, void *), void *arg)
|
|
{
|
|
sort_r (base, nmemb, size, compar, arg);
|
|
}
|
|
|
|
void *hc_bsearch_r (const void *key, const void *base, size_t nmemb, size_t size, int (*compar) (const void *, const void *, void *), void *arg)
|
|
{
|
|
for (size_t l = 0, r = nmemb; r; r >>= 1)
|
|
{
|
|
const size_t m = r >> 1;
|
|
|
|
const size_t c = l + m;
|
|
|
|
const char *next = (const char *) base + (c * size);
|
|
|
|
const int cmp = (*compar) (key, next, arg);
|
|
|
|
if (cmp > 0)
|
|
{
|
|
l += m + 1;
|
|
|
|
r--;
|
|
}
|
|
|
|
if (cmp == 0) return ((void *) next);
|
|
}
|
|
|
|
return (NULL);
|
|
}
|
|
|
|
bool hc_path_is_file (const char *path)
|
|
{
|
|
struct stat s;
|
|
|
|
memset (&s, 0, sizeof (s));
|
|
|
|
if (stat (path, &s) == -1) return false;
|
|
|
|
if (S_ISREG (s.st_mode)) return true;
|
|
|
|
return false;
|
|
}
|
|
|
|
bool hc_path_is_directory (const char *path)
|
|
{
|
|
struct stat s;
|
|
|
|
memset (&s, 0, sizeof (s));
|
|
|
|
if (stat (path, &s) == -1) return false;
|
|
|
|
if (S_ISDIR (s.st_mode)) return true;
|
|
|
|
return false;
|
|
}
|
|
|
|
bool hc_path_is_empty (const char *path)
|
|
{
|
|
struct stat s;
|
|
|
|
memset (&s, 0, sizeof (s));
|
|
|
|
if (stat (path, &s) == -1) return false;
|
|
|
|
if (s.st_size == 0) return true;
|
|
|
|
return false;
|
|
}
|
|
|
|
bool hc_path_exist (const char *path)
|
|
{
|
|
if (access (path, F_OK) == -1) return false;
|
|
|
|
return true;
|
|
}
|
|
|
|
bool hc_path_read (const char *path)
|
|
{
|
|
if (access (path, R_OK) == -1) return false;
|
|
|
|
return true;
|
|
}
|
|
|
|
bool hc_path_write (const char *path)
|
|
{
|
|
if (access (path, W_OK) == -1) return false;
|
|
|
|
return true;
|
|
}
|
|
|
|
bool hc_path_create (const char *path)
|
|
{
|
|
if (hc_path_exist (path) == true) return false;
|
|
|
|
#ifdef O_CLOEXEC
|
|
const int fd = open (path, O_WRONLY | O_CREAT | O_TRUNC | O_CLOEXEC, S_IRUSR | S_IWUSR);
|
|
#else
|
|
const int fd = open (path, O_WRONLY | O_CREAT | O_TRUNC, S_IRUSR | S_IWUSR);
|
|
#endif
|
|
|
|
if (fd == -1) return false;
|
|
|
|
close (fd);
|
|
|
|
unlink (path);
|
|
|
|
return true;
|
|
}
|
|
|
|
bool hc_path_has_bom (const char *path)
|
|
{
|
|
u8 buf[8] = { 0 };
|
|
|
|
HCFILE fp;
|
|
|
|
if (hc_fopen_raw (&fp, path, "rb") == false) return false;
|
|
|
|
const size_t nread = hc_fread (buf, 1, sizeof (buf), &fp);
|
|
|
|
hc_fclose (&fp);
|
|
|
|
if (nread < 1) return false;
|
|
|
|
const int bom_size = hc_string_bom_size (buf);
|
|
|
|
const bool has_bom = bom_size > 0;
|
|
|
|
return has_bom;
|
|
}
|
|
|
|
int hc_string_bom_size (const u8 *s)
|
|
{
|
|
/* signatures from https://en.wikipedia.org/wiki/Byte_order_mark#Byte_order_marks_by_encoding */
|
|
|
|
// utf-8
|
|
|
|
if ((s[0] == 0xef)
|
|
&& (s[1] == 0xbb)
|
|
&& (s[2] == 0xbf)) return 3;
|
|
|
|
// utf-16
|
|
|
|
if ((s[0] == 0xfe)
|
|
&& (s[1] == 0xff)) return 2;
|
|
|
|
if ((s[0] == 0xff)
|
|
&& (s[1] == 0xfe)) return 2;
|
|
|
|
// utf-32
|
|
|
|
if ((s[0] == 0x00)
|
|
&& (s[1] == 0x00)
|
|
&& (s[2] == 0xfe)
|
|
&& (s[3] == 0xff)) return 4;
|
|
|
|
if ((s[0] == 0xff)
|
|
&& (s[1] == 0xfe)
|
|
&& (s[2] == 0x00)
|
|
&& (s[3] == 0x00)) return 4;
|
|
|
|
// utf-7
|
|
|
|
if ((s[0] == 0x2b)
|
|
&& (s[1] == 0x2f)
|
|
&& (s[2] == 0x76)
|
|
&& (s[3] == 0x38)) return 4;
|
|
|
|
if ((s[0] == 0x2b)
|
|
&& (s[1] == 0x2f)
|
|
&& (s[2] == 0x76)
|
|
&& (s[3] == 0x39)) return 4;
|
|
|
|
if ((s[0] == 0x2b)
|
|
&& (s[1] == 0x2f)
|
|
&& (s[2] == 0x76)
|
|
&& (s[3] == 0x2b)) return 4;
|
|
|
|
if ((s[0] == 0x2b)
|
|
&& (s[1] == 0x2f)
|
|
&& (s[2] == 0x76)
|
|
&& (s[3] == 0x2f)) return 4;
|
|
|
|
if ((s[0] == 0x2b)
|
|
&& (s[1] == 0x2f)
|
|
&& (s[2] == 0x76)
|
|
&& (s[3] == 0x38)
|
|
&& (s[4] == 0x2d)) return 5;
|
|
|
|
// utf-1
|
|
|
|
if ((s[0] == 0xf7)
|
|
&& (s[1] == 0x64)
|
|
&& (s[2] == 0x4c)) return 3;
|
|
|
|
// utf-ebcdic
|
|
|
|
if ((s[0] == 0xdd)
|
|
&& (s[1] == 0x73)
|
|
&& (s[2] == 0x66)
|
|
&& (s[3] == 0x73)) return 4;
|
|
|
|
// scsu
|
|
|
|
if ((s[0] == 0x0e)
|
|
&& (s[1] == 0xfe)
|
|
&& (s[2] == 0xff)) return 3;
|
|
|
|
// bocu-1
|
|
|
|
if ((s[0] == 0xfb)
|
|
&& (s[1] == 0xee)
|
|
&& (s[2] == 0x28)) return 3;
|
|
|
|
// gb-18030
|
|
|
|
if ((s[0] == 0x84)
|
|
&& (s[1] == 0x31)
|
|
&& (s[2] == 0x95)
|
|
&& (s[3] == 0x33)) return 4;
|
|
|
|
return 0;
|
|
}
|
|
|
|
bool hc_string_is_digit (const char *s)
|
|
{
|
|
if (s == NULL) return false;
|
|
|
|
const size_t len = strlen (s);
|
|
|
|
if (len == 0) return false;
|
|
|
|
for (size_t i = 0; i < len; i++)
|
|
{
|
|
const int c = (const int) s[i];
|
|
|
|
if (isdigit (c) == 0) return false;
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
void setup_environment_variables (const folder_config_t *folder_config)
|
|
{
|
|
char *compute = getenv ("COMPUTE");
|
|
|
|
if (compute)
|
|
{
|
|
char *display;
|
|
|
|
hc_asprintf (&display, "DISPLAY=%s", compute);
|
|
|
|
putenv (display);
|
|
|
|
hcfree (display);
|
|
}
|
|
else
|
|
{
|
|
if (getenv ("DISPLAY") == NULL)
|
|
putenv ((char *) "DISPLAY=:0");
|
|
}
|
|
|
|
#if defined (DEBUG)
|
|
if (getenv ("OCL_CODE_CACHE_ENABLE") == NULL)
|
|
putenv ((char *) "OCL_CODE_CACHE_ENABLE=0");
|
|
|
|
if (getenv ("CUDA_CACHE_DISABLE") == NULL)
|
|
putenv ((char *) "CUDA_CACHE_DISABLE=1");
|
|
|
|
if (getenv ("POCL_KERNEL_CACHE") == NULL)
|
|
putenv ((char *) "POCL_KERNEL_CACHE=0");
|
|
#endif
|
|
|
|
if (getenv ("TMPDIR") == NULL)
|
|
{
|
|
char *tmpdir = NULL;
|
|
|
|
hc_asprintf (&tmpdir, "TMPDIR=%s", folder_config->profile_dir);
|
|
|
|
putenv (tmpdir);
|
|
|
|
// we can't free tmpdir at this point!
|
|
}
|
|
|
|
/*
|
|
if (getenv ("CL_CONFIG_USE_VECTORIZER") == NULL)
|
|
putenv ((char *) "CL_CONFIG_USE_VECTORIZER=False");
|
|
*/
|
|
|
|
#if defined (__CYGWIN__)
|
|
cygwin_internal (CW_SYNC_WINENV);
|
|
#endif
|
|
}
|
|
|
|
void setup_umask ()
|
|
{
|
|
umask (077);
|
|
}
|
|
|
|
void setup_seeding (const bool rp_gen_seed_chgd, const u32 rp_gen_seed)
|
|
{
|
|
if (rp_gen_seed_chgd == true)
|
|
{
|
|
srand (rp_gen_seed);
|
|
}
|
|
else
|
|
{
|
|
const time_t ts = time (NULL); // don't tell me that this is an insecure seed
|
|
|
|
srand ((unsigned int) ts);
|
|
}
|
|
}
|
|
|
|
u32 get_random_num (const u32 min, const u32 max)
|
|
{
|
|
if (min == max) return (min);
|
|
|
|
const u32 low = max - min;
|
|
|
|
if (low == 0) return (0);
|
|
|
|
#if defined (_WIN)
|
|
|
|
return (((u32) rand () % (max - min)) + min);
|
|
|
|
#else
|
|
|
|
return (((u32) random () % (max - min)) + min);
|
|
|
|
#endif
|
|
}
|
|
|
|
void hc_string_trim_leading (char *s)
|
|
{
|
|
int skip = 0;
|
|
|
|
const int len = (int) strlen (s);
|
|
|
|
for (int i = 0; i < len; i++)
|
|
{
|
|
const int c = (const int) s[i];
|
|
|
|
if (isspace (c) == 0) break;
|
|
|
|
skip++;
|
|
}
|
|
|
|
if (skip == 0) return;
|
|
|
|
const int new_len = len - skip;
|
|
|
|
memmove (s, s + skip, new_len);
|
|
|
|
s[new_len] = 0;
|
|
}
|
|
|
|
void hc_string_trim_trailing (char *s)
|
|
{
|
|
int skip = 0;
|
|
|
|
const int len = (int) strlen (s);
|
|
|
|
for (int i = len - 1; i >= 0; i--)
|
|
{
|
|
const int c = (const int) s[i];
|
|
|
|
if (isspace (c) == 0) break;
|
|
|
|
skip++;
|
|
}
|
|
|
|
if (skip == 0) return;
|
|
|
|
const size_t new_len = len - skip;
|
|
|
|
s[new_len] = 0;
|
|
}
|
|
|
|
int hc_get_processor_count ()
|
|
{
|
|
int cnt = 0;
|
|
|
|
#if defined (_WIN)
|
|
|
|
SYSTEM_INFO info;
|
|
|
|
GetSystemInfo (&info);
|
|
|
|
cnt = (int) info.dwNumberOfProcessors;
|
|
|
|
#else
|
|
|
|
cnt = (int) sysconf (_SC_NPROCESSORS_ONLN);
|
|
|
|
#endif
|
|
|
|
return cnt;
|
|
}
|
|
|
|
bool hc_same_files (char *file1, char *file2)
|
|
{
|
|
if ((file1 != NULL) && (file2 != NULL))
|
|
{
|
|
struct stat tmpstat_file1;
|
|
struct stat tmpstat_file2;
|
|
|
|
memset (&tmpstat_file1, 0, sizeof (tmpstat_file1));
|
|
memset (&tmpstat_file2, 0, sizeof (tmpstat_file2));
|
|
|
|
int do_check = 0;
|
|
|
|
HCFILE fp;
|
|
|
|
if (hc_fopen (&fp, file1, "r") == true)
|
|
{
|
|
if (hc_fstat (&fp, &tmpstat_file1))
|
|
{
|
|
hc_fclose (&fp);
|
|
|
|
return false;
|
|
}
|
|
|
|
hc_fclose (&fp);
|
|
|
|
do_check++;
|
|
}
|
|
|
|
if (hc_fopen (&fp, file2, "r") == true)
|
|
{
|
|
if (hc_fstat (&fp, &tmpstat_file2))
|
|
{
|
|
hc_fclose (&fp);
|
|
|
|
return false;
|
|
}
|
|
|
|
hc_fclose (&fp);
|
|
|
|
do_check++;
|
|
}
|
|
|
|
if (do_check == 2)
|
|
{
|
|
tmpstat_file1.st_mode = 0;
|
|
tmpstat_file1.st_nlink = 0;
|
|
tmpstat_file1.st_uid = 0;
|
|
tmpstat_file1.st_gid = 0;
|
|
tmpstat_file1.st_rdev = 0;
|
|
tmpstat_file1.st_atime = 0;
|
|
|
|
#if defined (STAT_NANOSECONDS_ACCESS_TIME)
|
|
tmpstat_file1.STAT_NANOSECONDS_ACCESS_TIME = 0;
|
|
#endif
|
|
|
|
#if defined (_POSIX)
|
|
tmpstat_file1.st_blksize = 0;
|
|
tmpstat_file1.st_blocks = 0;
|
|
#endif
|
|
|
|
tmpstat_file2.st_mode = 0;
|
|
tmpstat_file2.st_nlink = 0;
|
|
tmpstat_file2.st_uid = 0;
|
|
tmpstat_file2.st_gid = 0;
|
|
tmpstat_file2.st_rdev = 0;
|
|
tmpstat_file2.st_atime = 0;
|
|
|
|
#if defined (STAT_NANOSECONDS_ACCESS_TIME)
|
|
tmpstat_file2.STAT_NANOSECONDS_ACCESS_TIME = 0;
|
|
#endif
|
|
|
|
#if defined (_POSIX)
|
|
tmpstat_file2.st_blksize = 0;
|
|
tmpstat_file2.st_blocks = 0;
|
|
#endif
|
|
|
|
if (memcmp (&tmpstat_file1, &tmpstat_file2, sizeof (struct stat)) == 0)
|
|
{
|
|
return true;
|
|
}
|
|
}
|
|
}
|
|
|
|
return false;
|
|
}
|
|
|
|
u32 hc_strtoul (const char *nptr, char **endptr, int base)
|
|
{
|
|
return (u32) strtoul (nptr, endptr, base);
|
|
}
|
|
|
|
u64 hc_strtoull (const char *nptr, char **endptr, int base)
|
|
{
|
|
return (u64) strtoull (nptr, endptr, base);
|
|
}
|
|
|
|
u32 power_of_two_ceil_32 (const u32 v)
|
|
{
|
|
u32 r = v;
|
|
|
|
r--;
|
|
|
|
r |= r >> 1;
|
|
r |= r >> 2;
|
|
r |= r >> 4;
|
|
r |= r >> 8;
|
|
r |= r >> 16;
|
|
|
|
r++;
|
|
|
|
return r;
|
|
}
|
|
|
|
u32 power_of_two_floor_32 (const u32 v)
|
|
{
|
|
u32 r = power_of_two_ceil_32 (v);
|
|
|
|
if (r > v)
|
|
{
|
|
r >>= 1;
|
|
}
|
|
|
|
return r;
|
|
}
|
|
|
|
u32 round_up_multiple_32 (const u32 v, const u32 m)
|
|
{
|
|
if (m == 0) return v;
|
|
|
|
const u32 r = v % m;
|
|
|
|
if (r == 0) return v;
|
|
|
|
return v + m - r;
|
|
}
|
|
|
|
u64 round_up_multiple_64 (const u64 v, const u64 m)
|
|
{
|
|
if (m == 0) return v;
|
|
|
|
const u64 r = v % m;
|
|
|
|
if (r == 0) return v;
|
|
|
|
return v + m - r;
|
|
}
|
|
|
|
// difference to original strncat is no returncode and u8* instead of char*
|
|
|
|
void hc_strncat (u8 *dst, const u8 *src, const size_t n)
|
|
{
|
|
const size_t dst_len = strlen ((char *) dst);
|
|
|
|
const u8 *src_ptr = src;
|
|
|
|
u8 *dst_ptr = dst + dst_len;
|
|
|
|
for (size_t i = 0; i < n && *src_ptr != 0; i++)
|
|
{
|
|
*dst_ptr++ = *src_ptr++;
|
|
}
|
|
|
|
*dst_ptr = 0;
|
|
}
|
|
|
|
int count_char (const u8 *buf, const int len, const u8 c)
|
|
{
|
|
int r = 0;
|
|
|
|
for (int i = 0; i < len; i++)
|
|
{
|
|
if (buf[i] == c) r++;
|
|
}
|
|
|
|
return r;
|
|
}
|
|
|
|
float get_entropy (const u8 *buf, const int len)
|
|
{
|
|
float entropy = 0.0;
|
|
|
|
for (int c = 0; c < 256; c++)
|
|
{
|
|
const int r = count_char (buf, len, (const u8) c);
|
|
|
|
if (r == 0) continue;
|
|
|
|
float w = (float) r / len;
|
|
|
|
entropy += -w * log2f (w);
|
|
}
|
|
|
|
return entropy;
|
|
}
|
|
|
|
int select_read_timeout (int sockfd, const int sec)
|
|
{
|
|
struct timeval tv;
|
|
|
|
tv.tv_sec = sec;
|
|
tv.tv_usec = 0;
|
|
|
|
fd_set fds;
|
|
|
|
FD_ZERO (&fds);
|
|
FD_SET (sockfd, &fds);
|
|
|
|
return select (sockfd + 1, &fds, NULL, NULL, &tv);
|
|
}
|
|
|
|
int select_write_timeout (int sockfd, const int sec)
|
|
{
|
|
struct timeval tv;
|
|
|
|
tv.tv_sec = sec;
|
|
tv.tv_usec = 0;
|
|
|
|
fd_set fds;
|
|
|
|
FD_ZERO (&fds);
|
|
FD_SET (sockfd, &fds);
|
|
|
|
return select (sockfd + 1, NULL, &fds, NULL, &tv);
|
|
}
|
|
|
|
#if defined (_WIN)
|
|
|
|
int select_read_timeout_console (const int sec)
|
|
{
|
|
const HANDLE hStdIn = GetStdHandle (STD_INPUT_HANDLE);
|
|
|
|
const DWORD rc = WaitForSingleObject (hStdIn, sec * 1000);
|
|
|
|
if (rc == WAIT_OBJECT_0)
|
|
{
|
|
DWORD dwRead;
|
|
|
|
INPUT_RECORD inRecords;
|
|
|
|
inRecords.EventType = 0;
|
|
|
|
PeekConsoleInput (hStdIn, &inRecords, 1, &dwRead);
|
|
|
|
if (inRecords.EventType == 0)
|
|
{
|
|
// those are good ones
|
|
|
|
return 1;
|
|
}
|
|
else
|
|
{
|
|
// but we don't want that stuff like windows focus etc. in our stream
|
|
|
|
ReadConsoleInput (hStdIn, &inRecords, 1, &dwRead);
|
|
}
|
|
|
|
return select_read_timeout_console (sec);
|
|
}
|
|
else if (rc == WAIT_TIMEOUT)
|
|
{
|
|
return 0;
|
|
}
|
|
|
|
return -1;
|
|
}
|
|
|
|
#else
|
|
|
|
int select_read_timeout_console (const int sec)
|
|
{
|
|
return select_read_timeout (fileno (stdin), sec);
|
|
}
|
|
|
|
#endif
|
|
|
|
const char *strhashcategory (const u32 hash_category)
|
|
{
|
|
switch (hash_category)
|
|
{
|
|
case HASH_CATEGORY_UNDEFINED: return HASH_CATEGORY_UNDEFINED_STR;
|
|
case HASH_CATEGORY_RAW_HASH: return HASH_CATEGORY_RAW_HASH_STR;
|
|
case HASH_CATEGORY_RAW_HASH_SALTED: return HASH_CATEGORY_RAW_HASH_SALTED_STR;
|
|
case HASH_CATEGORY_RAW_HASH_AUTHENTICATED: return HASH_CATEGORY_RAW_HASH_AUTHENTICATED_STR;
|
|
case HASH_CATEGORY_RAW_CIPHER_KPA: return HASH_CATEGORY_RAW_CIPHER_KPA_STR;
|
|
case HASH_CATEGORY_GENERIC_KDF: return HASH_CATEGORY_GENERIC_KDF_STR;
|
|
case HASH_CATEGORY_NETWORK_PROTOCOL: return HASH_CATEGORY_NETWORK_PROTOCOL_STR;
|
|
case HASH_CATEGORY_FORUM_SOFTWARE: return HASH_CATEGORY_FORUM_SOFTWARE_STR;
|
|
case HASH_CATEGORY_DATABASE_SERVER: return HASH_CATEGORY_DATABASE_SERVER_STR;
|
|
case HASH_CATEGORY_NETWORK_SERVER: return HASH_CATEGORY_NETWORK_SERVER_STR;
|
|
case HASH_CATEGORY_RAW_CHECKSUM: return HASH_CATEGORY_RAW_CHECKSUM_STR;
|
|
case HASH_CATEGORY_OS: return HASH_CATEGORY_OS_STR;
|
|
case HASH_CATEGORY_EAS: return HASH_CATEGORY_EAS_STR;
|
|
case HASH_CATEGORY_ARCHIVE: return HASH_CATEGORY_ARCHIVE_STR;
|
|
case HASH_CATEGORY_FDE: return HASH_CATEGORY_FDE_STR;
|
|
case HASH_CATEGORY_FBE: return HASH_CATEGORY_FBE_STR;
|
|
case HASH_CATEGORY_DOCUMENTS: return HASH_CATEGORY_DOCUMENTS_STR;
|
|
case HASH_CATEGORY_PASSWORD_MANAGER: return HASH_CATEGORY_PASSWORD_MANAGER_STR;
|
|
case HASH_CATEGORY_OTP: return HASH_CATEGORY_OTP_STR;
|
|
case HASH_CATEGORY_PLAIN: return HASH_CATEGORY_PLAIN_STR;
|
|
case HASH_CATEGORY_FRAMEWORK: return HASH_CATEGORY_FRAMEWORK_STR;
|
|
case HASH_CATEGORY_PRIVATE_KEY: return HASH_CATEGORY_PRIVATE_KEY_STR;
|
|
case HASH_CATEGORY_IMS: return HASH_CATEGORY_IMS_STR;
|
|
case HASH_CATEGORY_CRYPTOCURRENCY_WALLET: return HASH_CATEGORY_CRYPTOCURRENCY_WALLET_STR;
|
|
}
|
|
|
|
return NULL;
|
|
}
|
|
|
|
const char *stroptitype (const u32 opti_type)
|
|
{
|
|
switch (opti_type)
|
|
{
|
|
case OPTI_TYPE_OPTIMIZED_KERNEL: return OPTI_STR_OPTIMIZED_KERNEL;
|
|
case OPTI_TYPE_ZERO_BYTE: return OPTI_STR_ZERO_BYTE;
|
|
case OPTI_TYPE_PRECOMPUTE_INIT: return OPTI_STR_PRECOMPUTE_INIT;
|
|
case OPTI_TYPE_MEET_IN_MIDDLE: return OPTI_STR_MEET_IN_MIDDLE;
|
|
case OPTI_TYPE_EARLY_SKIP: return OPTI_STR_EARLY_SKIP;
|
|
case OPTI_TYPE_NOT_SALTED: return OPTI_STR_NOT_SALTED;
|
|
case OPTI_TYPE_NOT_ITERATED: return OPTI_STR_NOT_ITERATED;
|
|
case OPTI_TYPE_PREPENDED_SALT: return OPTI_STR_PREPENDED_SALT;
|
|
case OPTI_TYPE_APPENDED_SALT: return OPTI_STR_APPENDED_SALT;
|
|
case OPTI_TYPE_SINGLE_HASH: return OPTI_STR_SINGLE_HASH;
|
|
case OPTI_TYPE_SINGLE_SALT: return OPTI_STR_SINGLE_SALT;
|
|
case OPTI_TYPE_BRUTE_FORCE: return OPTI_STR_BRUTE_FORCE;
|
|
case OPTI_TYPE_RAW_HASH: return OPTI_STR_RAW_HASH;
|
|
case OPTI_TYPE_SLOW_HASH_SIMD_INIT: return OPTI_STR_SLOW_HASH_SIMD_INIT;
|
|
case OPTI_TYPE_SLOW_HASH_SIMD_LOOP: return OPTI_STR_SLOW_HASH_SIMD_LOOP;
|
|
case OPTI_TYPE_SLOW_HASH_SIMD_COMP: return OPTI_STR_SLOW_HASH_SIMD_COMP;
|
|
case OPTI_TYPE_USES_BITS_8: return OPTI_STR_USES_BITS_8;
|
|
case OPTI_TYPE_USES_BITS_16: return OPTI_STR_USES_BITS_16;
|
|
case OPTI_TYPE_USES_BITS_32: return OPTI_STR_USES_BITS_32;
|
|
case OPTI_TYPE_USES_BITS_64: return OPTI_STR_USES_BITS_64;
|
|
}
|
|
|
|
return NULL;
|
|
}
|
|
|
|
const char *strparser (const u32 parser_status)
|
|
{
|
|
switch (parser_status)
|
|
{
|
|
case PARSER_OK: return PA_000;
|
|
case PARSER_COMMENT: return PA_001;
|
|
case PARSER_GLOBAL_ZERO: return PA_002;
|
|
case PARSER_GLOBAL_LENGTH: return PA_003;
|
|
case PARSER_HASH_LENGTH: return PA_004;
|
|
case PARSER_HASH_VALUE: return PA_005;
|
|
case PARSER_SALT_LENGTH: return PA_006;
|
|
case PARSER_SALT_VALUE: return PA_007;
|
|
case PARSER_SALT_ITERATION: return PA_008;
|
|
case PARSER_SEPARATOR_UNMATCHED: return PA_009;
|
|
case PARSER_SIGNATURE_UNMATCHED: return PA_010;
|
|
case PARSER_HCCAPX_FILE_SIZE: return PA_011;
|
|
case PARSER_HCCAPX_EAPOL_LEN: return PA_012;
|
|
case PARSER_PSAFE2_FILE_SIZE: return PA_013;
|
|
case PARSER_PSAFE3_FILE_SIZE: return PA_014;
|
|
case PARSER_TC_FILE_SIZE: return PA_015;
|
|
case PARSER_VC_FILE_SIZE: return PA_016;
|
|
case PARSER_SIP_AUTH_DIRECTIVE: return PA_017;
|
|
case PARSER_HASH_FILE: return PA_018;
|
|
case PARSER_HASH_ENCODING: return PA_019;
|
|
case PARSER_SALT_ENCODING: return PA_020;
|
|
case PARSER_LUKS_FILE_SIZE: return PA_021;
|
|
case PARSER_LUKS_MAGIC: return PA_022;
|
|
case PARSER_LUKS_VERSION: return PA_023;
|
|
case PARSER_LUKS_CIPHER_TYPE: return PA_024;
|
|
case PARSER_LUKS_CIPHER_MODE: return PA_025;
|
|
case PARSER_LUKS_HASH_TYPE: return PA_026;
|
|
case PARSER_LUKS_KEY_SIZE: return PA_027;
|
|
case PARSER_LUKS_KEY_DISABLED: return PA_028;
|
|
case PARSER_LUKS_KEY_STRIPES: return PA_029;
|
|
case PARSER_LUKS_HASH_CIPHER: return PA_030;
|
|
case PARSER_HCCAPX_SIGNATURE: return PA_031;
|
|
case PARSER_HCCAPX_VERSION: return PA_032;
|
|
case PARSER_HCCAPX_MESSAGE_PAIR: return PA_033;
|
|
case PARSER_TOKEN_ENCODING: return PA_034;
|
|
case PARSER_TOKEN_LENGTH: return PA_035;
|
|
case PARSER_INSUFFICIENT_ENTROPY: return PA_036;
|
|
case PARSER_PKZIP_CT_UNMATCHED: return PA_037;
|
|
case PARSER_KEY_SIZE: return PA_038;
|
|
case PARSER_BLOCK_SIZE: return PA_039;
|
|
case PARSER_CIPHER: return PA_040;
|
|
case PARSER_FILE_SIZE: return PA_041;
|
|
case PARSER_IV_LENGTH: return PA_042;
|
|
case PARSER_CT_LENGTH: return PA_043;
|
|
}
|
|
|
|
return PA_255;
|
|
}
|
|
|
|
static int rounds_count_length (const char *input_buf, const int input_len)
|
|
{
|
|
if (input_len >= 9) // 9 is minimum because of "rounds=X$"
|
|
{
|
|
static const char *const rounds = "rounds=";
|
|
|
|
if (memcmp (input_buf, rounds, 7) == 0)
|
|
{
|
|
const char *next_pos = strchr (input_buf + 8, '$');
|
|
|
|
if (next_pos == NULL) return -1;
|
|
|
|
const int rounds_len = next_pos - input_buf;
|
|
|
|
return rounds_len;
|
|
}
|
|
}
|
|
|
|
return -1;
|
|
}
|
|
|
|
const u8 *hc_strchr_next (const u8 *input_buf, const int input_len, const u8 separator)
|
|
{
|
|
for (int i = 0; i < input_len; i++)
|
|
{
|
|
if (input_buf[i] == separator) return &input_buf[i];
|
|
}
|
|
|
|
return NULL;
|
|
}
|
|
|
|
const u8 *hc_strchr_last (const u8 *input_buf, const int input_len, const u8 separator)
|
|
{
|
|
for (int i = input_len - 1; i >= 0; i--)
|
|
{
|
|
if (input_buf[i] == separator) return &input_buf[i];
|
|
}
|
|
|
|
return NULL;
|
|
}
|
|
|
|
int input_tokenizer (const u8 *input_buf, const int input_len, token_t *token)
|
|
{
|
|
int len_left = input_len;
|
|
|
|
token->buf[0] = input_buf;
|
|
|
|
int token_idx;
|
|
|
|
for (token_idx = 0; token_idx < token->token_cnt - 1; token_idx++)
|
|
{
|
|
if (token->attr[token_idx] & TOKEN_ATTR_FIXED_LENGTH)
|
|
{
|
|
int len = token->len[token_idx];
|
|
|
|
if (len_left < len) return (PARSER_TOKEN_LENGTH);
|
|
|
|
token->buf[token_idx + 1] = token->buf[token_idx] + len;
|
|
|
|
len_left -= len;
|
|
}
|
|
else
|
|
{
|
|
if (token->attr[token_idx] & TOKEN_ATTR_OPTIONAL_ROUNDS)
|
|
{
|
|
const int len = rounds_count_length ((const char *) token->buf[token_idx], len_left);
|
|
|
|
token->opt_buf = token->buf[token_idx];
|
|
|
|
token->opt_len = len; // we want an eventual -1 in here, it's used later for verification
|
|
|
|
if (len > 0)
|
|
{
|
|
token->buf[token_idx] += len + 1; // +1 = separator
|
|
|
|
len_left -= len + 1; // +1 = separator
|
|
}
|
|
}
|
|
|
|
const u8 *next_pos = NULL;
|
|
|
|
if (token->attr[token_idx] & TOKEN_ATTR_SEPARATOR_FARTHEST)
|
|
{
|
|
next_pos = hc_strchr_last (token->buf[token_idx], len_left, token->sep[token_idx]);
|
|
}
|
|
else
|
|
{
|
|
next_pos = hc_strchr_next (token->buf[token_idx], len_left, token->sep[token_idx]);
|
|
}
|
|
|
|
if (next_pos == NULL) return (PARSER_SEPARATOR_UNMATCHED);
|
|
|
|
const int len = next_pos - token->buf[token_idx];
|
|
|
|
token->len[token_idx] = len;
|
|
|
|
token->buf[token_idx + 1] = next_pos + 1; // +1 = separator
|
|
|
|
len_left -= len + 1; // +1 = separator
|
|
}
|
|
}
|
|
|
|
if (token->attr[token_idx] & TOKEN_ATTR_FIXED_LENGTH)
|
|
{
|
|
int len = token->len[token_idx];
|
|
|
|
if (len_left != len) return (PARSER_TOKEN_LENGTH);
|
|
}
|
|
else
|
|
{
|
|
token->len[token_idx] = len_left;
|
|
}
|
|
|
|
// verify data
|
|
|
|
for (token_idx = 0; token_idx < token->token_cnt; token_idx++)
|
|
{
|
|
if (token->attr[token_idx] & TOKEN_ATTR_VERIFY_SIGNATURE)
|
|
{
|
|
bool matched = false;
|
|
|
|
for (int signature_idx = 0; signature_idx < token->signatures_cnt; signature_idx++)
|
|
{
|
|
if (strncmp ((char *) token->buf[token_idx], token->signatures_buf[signature_idx], token->len[token_idx]) == 0) matched = true;
|
|
}
|
|
|
|
if (matched == false) return (PARSER_SIGNATURE_UNMATCHED);
|
|
}
|
|
|
|
if (token->attr[token_idx] & TOKEN_ATTR_VERIFY_LENGTH)
|
|
{
|
|
if (token->len[token_idx] < token->len_min[token_idx]) return (PARSER_TOKEN_LENGTH);
|
|
if (token->len[token_idx] > token->len_max[token_idx]) return (PARSER_TOKEN_LENGTH);
|
|
}
|
|
|
|
if (token->attr[token_idx] & TOKEN_ATTR_VERIFY_DIGIT)
|
|
{
|
|
if (is_valid_digit_string (token->buf[token_idx], token->len[token_idx]) == false) return (PARSER_TOKEN_ENCODING);
|
|
}
|
|
|
|
if (token->attr[token_idx] & TOKEN_ATTR_VERIFY_FLOAT)
|
|
{
|
|
if (is_valid_float_string (token->buf[token_idx], token->len[token_idx]) == false) return (PARSER_TOKEN_ENCODING);
|
|
}
|
|
|
|
if (token->attr[token_idx] & TOKEN_ATTR_VERIFY_HEX)
|
|
{
|
|
if (is_valid_hex_string (token->buf[token_idx], token->len[token_idx]) == false) return (PARSER_TOKEN_ENCODING);
|
|
}
|
|
|
|
if (token->attr[token_idx] & TOKEN_ATTR_VERIFY_BASE64A)
|
|
{
|
|
if (is_valid_base64a_string (token->buf[token_idx], token->len[token_idx]) == false) return (PARSER_TOKEN_ENCODING);
|
|
}
|
|
|
|
if (token->attr[token_idx] & TOKEN_ATTR_VERIFY_BASE64B)
|
|
{
|
|
if (is_valid_base64b_string (token->buf[token_idx], token->len[token_idx]) == false) return (PARSER_TOKEN_ENCODING);
|
|
}
|
|
|
|
if (token->attr[token_idx] & TOKEN_ATTR_VERIFY_BASE64C)
|
|
{
|
|
if (is_valid_base64c_string (token->buf[token_idx], token->len[token_idx]) == false) return (PARSER_TOKEN_ENCODING);
|
|
}
|
|
}
|
|
|
|
return PARSER_OK;
|
|
}
|
|
|
|
bool generic_salt_decode (MAYBE_UNUSED const hashconfig_t *hashconfig, const u8 *in_buf, const int in_len, u8 *out_buf, int *out_len)
|
|
{
|
|
u32 tmp_u32[(64 * 2) + 1] = { 0 };
|
|
|
|
u8 *tmp_u8 = (u8 *) tmp_u32;
|
|
|
|
if (in_len > 512) return false; // 512 = 2 * 256 -- (2 * because of hex), 256 because of maximum salt length in salt_t
|
|
|
|
int tmp_len = 0;
|
|
|
|
if (hashconfig->opts_type & OPTS_TYPE_ST_HEX)
|
|
{
|
|
if (in_len < (int) (hashconfig->salt_min * 2)) return false;
|
|
if (in_len > (int) (hashconfig->salt_max * 2)) return false;
|
|
|
|
if (in_len & 1) return false;
|
|
|
|
for (int i = 0, j = 0; j < in_len; i += 1, j += 2)
|
|
{
|
|
u8 p0 = in_buf[j + 0];
|
|
u8 p1 = in_buf[j + 1];
|
|
|
|
tmp_u8[i] = hex_convert (p1) << 0;
|
|
tmp_u8[i] |= hex_convert (p0) << 4;
|
|
}
|
|
|
|
tmp_len = in_len / 2;
|
|
}
|
|
else if (hashconfig->opts_type & OPTS_TYPE_ST_BASE64)
|
|
{
|
|
if (in_len < (int) (((hashconfig->salt_min * 8) / 6) + 0)) return false;
|
|
if (in_len > (int) (((hashconfig->salt_max * 8) / 6) + 3)) return false;
|
|
|
|
tmp_len = base64_decode (base64_to_int, in_buf, in_len, tmp_u8);
|
|
}
|
|
else
|
|
{
|
|
if (in_len < (int) hashconfig->salt_min) return false;
|
|
if (in_len > (int) hashconfig->salt_max) return false;
|
|
|
|
memcpy (tmp_u8, in_buf, in_len);
|
|
|
|
tmp_len = in_len;
|
|
}
|
|
|
|
if (hashconfig->opts_type & OPTS_TYPE_ST_UTF16LE)
|
|
{
|
|
if (tmp_len >= 128) return false;
|
|
|
|
for (int i = 64 - 1; i >= 1; i -= 2)
|
|
{
|
|
const u32 v = tmp_u32[i / 2];
|
|
|
|
tmp_u32[i - 0] = ((v >> 8) & 0x00FF0000) | ((v >> 16) & 0x000000FF);
|
|
tmp_u32[i - 1] = ((v << 8) & 0x00FF0000) | ((v >> 0) & 0x000000FF);
|
|
}
|
|
|
|
tmp_len = tmp_len * 2;
|
|
}
|
|
|
|
if (hashconfig->opts_type & OPTS_TYPE_ST_LOWER)
|
|
{
|
|
lowercase (tmp_u8, tmp_len);
|
|
}
|
|
|
|
if (hashconfig->opts_type & OPTS_TYPE_ST_UPPER)
|
|
{
|
|
uppercase (tmp_u8, tmp_len);
|
|
}
|
|
|
|
int tmp2_len = tmp_len;
|
|
|
|
if (hashconfig->opts_type & OPTS_TYPE_ST_ADD80)
|
|
{
|
|
if (tmp2_len >= 256) return false;
|
|
|
|
tmp_u8[tmp2_len++] = 0x80;
|
|
}
|
|
|
|
if (hashconfig->opts_type & OPTS_TYPE_ST_ADD01)
|
|
{
|
|
if (tmp2_len >= 256) return false;
|
|
|
|
tmp_u8[tmp2_len++] = 0x01;
|
|
}
|
|
|
|
memcpy (out_buf, tmp_u8, tmp2_len);
|
|
|
|
*out_len = tmp_len;
|
|
|
|
return true;
|
|
}
|
|
|
|
int generic_salt_encode (MAYBE_UNUSED const hashconfig_t *hashconfig, const u8 *in_buf, const int in_len, u8 *out_buf)
|
|
{
|
|
u32 tmp_u32[(64 * 2) + 1] = { 0 };
|
|
|
|
u8 *tmp_u8 = (u8 *) tmp_u32;
|
|
|
|
memcpy (tmp_u8, in_buf, in_len);
|
|
|
|
int tmp_len = in_len;
|
|
|
|
if (hashconfig->opts_type & OPTS_TYPE_ST_UTF16LE)
|
|
{
|
|
for (int i = 0, j = 0; j < in_len; i += 1, j += 2)
|
|
{
|
|
const u8 p = tmp_u8[j];
|
|
|
|
tmp_u8[i] = p;
|
|
}
|
|
|
|
tmp_len = tmp_len / 2;
|
|
}
|
|
|
|
if (hashconfig->opts_type & OPTS_TYPE_ST_HEX)
|
|
{
|
|
for (int i = 0, j = 0; i < in_len; i += 1, j += 2)
|
|
{
|
|
u8_to_hex (in_buf[i], tmp_u8 + j);
|
|
}
|
|
|
|
tmp_len = in_len * 2;
|
|
}
|
|
else if (hashconfig->opts_type & OPTS_TYPE_ST_BASE64)
|
|
{
|
|
tmp_len = base64_encode (int_to_base64, in_buf, in_len, tmp_u8);
|
|
}
|
|
|
|
memcpy (out_buf, tmp_u8, tmp_len);
|
|
|
|
return tmp_len;
|
|
}
|