Cryptography Pitfalls at ConFoo Montreal 2017

Cryptography Pitfalls at ConFoo Montreal 2017

58376779023f009fc13d160bb3e82515?s=128

John Downey

March 10, 2017
Tweet

Transcript

  1. Cryptography Pitfalls John Downey | @jtdowney @jtdowney 1

  2. @jtdowney 2

  3. @jtdowney 3

  4. @jtdowney 4

  5. The views expressed in this presentation are my own, and

    not those of PayPal or any of its affiliates. @jtdowney 5
  6. @jtdowney 6

  7. Confidentiality @jtdowney 7

  8. Authentication @jtdowney 8

  9. Identification @jtdowney 9

  10. Rigorous Science @jtdowney 10

  11. Peer Review @jtdowney 11

  12. @jtdowney 12

  13. You have probably seen the door to a bank vault,

    at least in the movies. You know, 10-inch- thick, hardened steel, with huge bolts to lock it in place. It certainly looks impressive. We often find the digital equivalent of such a vault door installed in a tent. The people standing around it are arguing over how thick the door should be, rather than spending their time looking at the tent. — Cryptography Engineering by Niels Ferguson, Bruce Schneier, and Tadayoshi Kohno @jtdowney 13
  14. • For data in transit • Use TLS (née SSL),

    SSH, or VPN/IPsec • For data at rest • Use GnuPG • Data to be signed • Use GnuPG @jtdowney 14
  15. • Avoid low level libraries • OpenSSL • PyCrypto •

    Bouncy Castle • Use a high level library • NaCL/libsodium (C, Ruby, PHP, etc) • Keyczar (C++, Python, and Java) @jtdowney 15
  16. @jtdowney 16

  17. Random Number Generation @jtdowney 17

  18. Pitfalls 1. Not using a cryptographically strong random number generator

    2. Not using random data when it is required 3. Broken random number generators @jtdowney 18
  19. @jtdowney 19

  20. @jtdowney 20

  21. Pitfalls 1. Not using a cryptographically strong random number generator

    2. Not using random data when it is required 3. Broken random number generators @jtdowney 21
  22. @jtdowney 22

  23. Pitfalls 1. Not using a cryptographically strong random number generator

    2. Not using random data when it is required 3. Broken random number generators @jtdowney 23
  24. @jtdowney 24

  25. @jtdowney 25

  26. MD_Update(&m,buf,j); @jtdowney 26

  27. Don't add uninitialised data to the random number generator. This

    stop valgrind from giving error messages in unrelated code. (Closes: #363516) @jtdowney 27
  28. /* DO NOT REMOVE THE FOLLOWING CALL TO MD_Update()! */

    MD_Update(&m,buf,j); /* We know that line may cause programs such as purify and valgrind to complain about use of uninitialized data. The problem is not, it's with the caller. Removing that line will make sure you get really bad randomness and thereby other problems such as very insecure keys. */ @jtdowney 28
  29. @jtdowney 29

  30. @jtdowney 30

  31. Recommendations • Use a cryptographically strong random number generator •

    Unix-like • Read from /dev/urandom • Windows • RandomNumberGenerator in System.Security.Cryptography (.NET) • CryptGenRandom • Java use java.security.SecureRandom @jtdowney 31
  32. Hash Functions @jtdowney 32

  33. Pitfalls 1. Using weak/old algorithms 2. Misunderstanding checksums 3. Length

    extension attacks @jtdowney 33
  34. @jtdowney 34

  35. @jtdowney 35

  36. @jtdowney 36

  37. @jtdowney 37

  38. 9EC4C12949A4F31474F299058CE2B22A @jtdowney 38

  39. 9EC4C12949A4F31474F299058CE2B22A USCYBERCOM plans, coordinates, integrates, synchronizes and conducts activities to:

    direct the operations and defense of specified Department of Defense information networks and; prepare to, and when directed, conduct full spectrum military cyberspace operations in order to enable actions in all domains, ensure US/Allied freedom of action in cyberspace and deny the same to our adversaries. @jtdowney 39
  40. Pitfalls 1. Using weak/old algorithms 2. Misunderstanding checksums 3. Length

    extension attacks @jtdowney 40
  41. @jtdowney 41

  42. Pitfalls 1. Using weak/old algorithms 2. Misunderstanding checksums 3. Length

    extension attacks @jtdowney 42
  43. Message Authentication Code (MAC) tag = MAC(key, value) • Takes:

    • key - shared secret • value - value to protected integrity of • Returns: • tag - value that represents the integrity @jtdowney 43
  44. Naive approach tag = sha256(key + value) @jtdowney 44

  45. Length Extension Attacks secret = "my-secret-key" value = "buy 10

    units at $1" signature = sha256(secret + value) @jtdowney 45
  46. Length Extension Attacks secret = "my-secret-key" value = "buy 10

    units at $1" + " or $0" signature = sha256(secret + value) @jtdowney 46
  47. Fixed secret = "my-secret-key" value = "buy 10 units at

    $1" signature = hmac_sha256(secret, value) @jtdowney 47
  48. @jtdowney 48

  49. @jtdowney 49

  50. Recommendations • Use SHA-256 (SHA-2 family) • Choose HMAC-SHA-256 if

    you want a signature • Use BLAKE2b if you need speed • Stop using MD5 • Stop using SHA1 @jtdowney 50
  51. Passwords @jtdowney 51

  52. @jtdowney 52

  53. @jtdowney 53

  54. @jtdowney 54

  55. @jtdowney 55

  56. @jtdowney 56

  57. @jtdowney 57

  58. @jtdowney 58

  59. @jtdowney 59

  60. @jtdowney 60

  61. @jtdowney 61

  62. Hash functions are fast @jtdowney 62

  63. 1. One-way • Value can be used for verification 2.

    Randomized • Can largely defeat pre-computed tables • Forces attackers to focus on one password 3. Slow @jtdowney 63
  64. Adaptive Hashing bcrypt, scrypt, argon2, or PBKDF2 @jtdowney 64

  65. Recommendations • Delegate authentication if possible • Facebook, Twitter, Google,

    Github • Store one-way verifiers using bcrypt, scrypt, argon2, or PBDKF2 @jtdowney 65
  66. So your password storage is bad @jtdowney 66

  67. It will be ok, you can fix it @jtdowney 67

  68. Example: password_hash column is sha1(salt + password) @jtdowney 68

  69. • Don't wait for user to login and silently upgrade

    • Wrap bcrypt around existing scheme • Use bcrypt(sha1(salt + password) • Upgrade all passwords in place • This does require the previous password scheme wasn't atrociously bad (e.g. DES crypt) @jtdowney 69
  70. Now: password_hash column is bcrypt(sha1(salt + password)) @jtdowney 70

  71. Ciphers @jtdowney 71

  72. Pitfalls 1. Using old/weak algorithms 2. Using ECB mode for

    block ciphers 3. Not using authenticated encryption @jtdowney 72
  73. @jtdowney 73

  74. @jtdowney 74

  75. @jtdowney 75

  76. Pitfalls 1. Using old/weak algorithms 2. Using ECB mode for

    block ciphers 3. Not using authenticated encryption @jtdowney 76
  77. AES - primitive ciphertext = AES_Encrypt(key, plaintext) plaintext = AES_Decrypt(key,

    ciphertext) • Function over: • key - 128, 192, or 256 bit value • plaintext - 128 bit value • ciphertext - 128 bit value @jtdowney 77
  78. ECB Encrypt while (remaining blocks) { block = ... #

    next 16 byte (128 bit chunk) ouput.append(AES_Encrypt(key, block)) } @jtdowney 78
  79. @jtdowney 79

  80. @jtdowney 80

  81. Pitfalls 1. Using old/weak algorithms 2. Using ECB mode for

    block ciphers 3. Not using authenticated encryption @jtdowney 81
  82. @jtdowney 82

  83. @jtdowney 83

  84. @jtdowney 84

  85. @jtdowney 85

  86. World of hurt @jtdowney 86

  87. Recommendations • Prefer to use box/secret box from NaCL/ libsodium

    • Stop using DES • Stop building your own on top of AES • Stop encrypting without protecting integrity @jtdowney 87
  88. What if you have to use AES • Do not

    use ECB mode • Be sure to use authenticated encryption • GCM mode would be a good first choice • Verify the tag/MAC first • Still easy to mess up in a critical way @jtdowney 88
  89. TLS/SSL @jtdowney 89

  90. Pitfalls 1. Not verifying the certificate chain or hostname 2.

    Misconfigured server settings 3. Using a broken library @jtdowney 90
  91. @jtdowney 91

  92. @jtdowney 92

  93. @jtdowney 93

  94. @jtdowney 94

  95. Hostname verification @jtdowney 95

  96. Hostname verification • Check that you got the certificate for

    who you intended to connect to • Hostname verification is protocol dependent • OpenSSL doesn't have it built in @jtdowney 96
  97. Pitfalls 1. Not verifying the certificate chain or hostname 2.

    Misconfigured server settings 3. Using a broken library @jtdowney 97
  98. @jtdowney 98

  99. SSL Labs https://www.ssllabs.com @jtdowney 99

  100. testssl.sh https://testssl.sh @jtdowney 100

  101. TLS Server Settings https://mozilla.github.io/server-side-tls/ssl-config-generator/ @jtdowney 101

  102. Pitfalls 1. Not verifying the certificate chain or hostname 2.

    Misconfigured server settings 3. Using a broken library @jtdowney 102
  103. @jtdowney 103

  104. @jtdowney 104

  105. Recommendations • Do ensure you're validating connections • Lean on

    a framework/library if possible • But check that it also does the right thing • Setup and automated test to validate this setting (badssl.com) @jtdowney 105
  106. Trust @jtdowney 106

  107. The authenticity of host 'apollo.local (10.0.2.56)' can't be established. RSA

    key fingerprint is 04:63:c1:ba:c7:31:04:12:14:ff:b6:c4:32:cf:44:ec. Are you sure you want to continue connecting (yes/no)? @jtdowney 107
  108. @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ @ WARNING: REMOTE HOST IDENTIFICATION HAS CHANGED! @ @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@

    IT IS POSSIBLE THAT SOMEONE IS DOING SOMETHING NASTY! Someone could be eavesdropping on you right now (man-in-the-middle attack)! It is also possible that the RSA host key has just been changed. The fingerprint for the RSA key sent by the remote host is 04:63:c1:ba:c7:31:04:12:14:ff:b6:c4:32:cf:44:ec. Please contact your system administrator. @jtdowney 108
  109. @jtdowney 109

  110. Certificate Pinning @jtdowney 110

  111. @jtdowney 111

  112. Recommendations • Think about what organizations you really trust •

    Investigate certificate pinning for your apps @jtdowney 112
  113. @jtdowney 113

  114. Stanford Crypto Class http://crypto-class.com @jtdowney 114

  115. Matasano Crypto Challenges http://cryptopals.com @jtdowney 115

  116. Questions John Downey | @jtdowney @jtdowney 116

  117. Bonus Round @jtdowney 117

  118. Quantum Computers @jtdowney 118

  119. Pitfalls 1. Assuming current crypto will last forever @jtdowney 119

  120. @jtdowney 120

  121. @jtdowney 121

  122. Recommendations • Follow the PQCrypto discussion • Stay away from

    PQCrypto until the industry starts to standardize • Hope that researchers are moving fast enough @jtdowney 122
  123. Images • https://flic.kr/p/6eagaw • https://flic.kr/p/4KWhKn • https://flic.kr/p/9F2BCv • https://flic.kr/p/486xYS •

    https://flic.kr/p/7Ffppm • https://flic.kr/p/8TuJD9 • https://flic.kr/p/4iLJZt • https://flic.kr/p/4pGZuz • https://flic.kr/p/48w7wP • https://flic.kr/p/8aZWNE • https://flic.kr/p/5NRHp • https://flic.kr/p/7p7raq • https://flic.kr/p/aZEE1Z • https://flic.kr/p/7WtwAz • https://flic.kr/p/6AN9mM • https://flic.kr/p/6dt62u • https://flic.kr/p/4ZqwyB • https://flic.kr/p/Bqewr • https://flic.kr/p/ecdhVE • https://flic.kr/p/AV1Nd • https://flic.kr/p/5tWgh4 @jtdowney 123