| I use scrypt[1] as a component of http://masterpasswordapp.com, a stateless GPL password "manager" which does away with password vaults and generates site passwords from memorable tokens only. My usage of scrypt[2] involves little more than: crypto_scrypt( (const uint8_t *)masterPassword, strlen(masterPassword), (const uint8_t *)masterKeySalt, masterKeySaltLength, MP_N, MP_r, MP_p, masterKey, MP_dkLen ) Where: - masterPassword is a C-string read from a file (file was not touched during these tests) - masterKeySalt is determined based on external factors (was logged as constant during these tests) - MP_N, MP_r, MP_p, MP_dkLen are constants defined as 32768, 8, 2 and 64 respectively. This function always produces identical output, with the sole exception of a single instance I've seen where it produced different output. This was logged as such: [EXPECTED:] userName: Robert Lee Mitchell mpwConfigPath: /Users/lhunath/.mpw masterPassword: banana colored duckling Robert Lee Mitchell's password for ⛄: key scope: com.lyndir.masterpassword masterKeySalt ID: D006ACE8060B83938FE812CC1C20D74F4EE9F6F33C3A1BC48E6CC3C43BCDD541 masterKey ID: 98EEF4D1DF46D849574A82A03C3177056B15DFFCA29BB3899DE4628453675302 (derived in 0.20s) [UNEXPECTED:] userName: Robert Lee Mitchell mpwConfigPath: /Users/lhunath/.mpw masterPassword: banana colored duckling Robert Lee Mitchell's password for ⛄: [ ╚☻╯⛄ ]: key scope: com.lyndir.masterpassword masterKeySalt ID: D006ACE8060B83938FE812CC1C20D74F4EE9F6F33C3A1BC48E6CC3C43BCDD541 masterKey ID: 4312E1DEB1BEC1305AC2DE3F366F91F8B39D130027206370FE2E23931C6920F1 (derived in 0.19s) Where the "ID" output is the result of performing your SHA256_Buf on the buffer pointer and then formatting it as hexadecimal output using: const char *IDForBuf(const void *buf, size_t length) { uint8_t hash[32]; SHA256_Buf(buf, length, hash); char *id = (char *)calloc(65, sizeof(char)); for (int kH = 0; kH < 32; kH++) sprintf(&(id[kH * 2]), "%02X", hash[kH]); return id; } My question: Is there a theory that might explain this, or could there be a bug in the scrypt implementation used. FWIW, this machine does not have ECC RAM. [1] scrypt implementation used: https://code.google.com/p/scrypt/ [2] my usage of it: https://github.com/Lyndir/MasterPassword/blob/master/MasterPassword/C/mpw.c#L249 — Maarten Billemont (lhunath) — me: http://www.lhunath.com – business: http://www.lyndir.com – http://masterpasswordapp.com
|
Attachment:
smime.p7s
Description: S/MIME cryptographic signature