| | | 1 | | using System.Security.Cryptography; |
| | | 2 | | using System.Text; |
| | | 3 | | |
| | | 4 | | namespace ArturRios.Heimdall.Command.Services; |
| | | 5 | | |
| | | 6 | | /// <summary>A recovery code as issued: the form shown to the person once, and the digest stored for it.</summary> |
| | | 7 | | /// <param name="Code">The code as displayed, <c>XXXX-XXXX-XXXX-XXXX</c>.</param> |
| | | 8 | | /// <param name="CodeHash">What <c>TwoFactorRecoveryCode.CodeHash</c> stores for it.</param> |
| | | 9 | | public sealed record IssuedRecoveryCode(string Code, byte[] CodeHash); |
| | | 10 | | |
| | | 11 | | /// <summary> |
| | | 12 | | /// The second-factor recovery codes of FR-2F-05 and FR-2F-12: how they are generated, how a |
| | | 13 | | /// presented one is normalised, and how both are hashed. UC-37 and UC-40 issue them and |
| | | 14 | | /// <see cref="TwoFactorFactorVerifier" /> matches them, and all three go through here so the |
| | | 15 | | /// value written at issue and the value compared at presentation cannot drift apart. |
| | | 16 | | /// </summary> |
| | | 17 | | /// <remarks> |
| | | 18 | | /// <para> |
| | | 19 | | /// <b>Shape.</b> Ten random bytes from <see cref="RandomNumberGenerator" />, 80 bits, written |
| | | 20 | | /// as sixteen characters of Crockford's base32 (<see cref="Alphabet" />: the digits and the |
| | | 21 | | /// uppercase letters without I, L, O and U) and shown in four groups of four, |
| | | 22 | | /// <c>XXXX-XXXX-XXXX-XXXX</c>. Crockford's rather than RFC 4648's because it leaves out the |
| | | 23 | | /// letters people misread for digits and reads them back as those digits instead: a code |
| | | 24 | | /// typed with an O, an I or an L still matches. |
| | | 25 | | /// </para> |
| | | 26 | | /// <para> |
| | | 27 | | /// <b>What a presented code may look like.</b> Hyphens and whitespace are ignored wherever |
| | | 28 | | /// they fall, and case is ignored. What remains is sixteen characters for a code issued |
| | | 29 | | /// now, or eight for one issued before (two groups of four from <c>[A-Z0-9]</c>, stored as |
| | | 30 | | /// the digest of the exact <c>XXXX-XXXX</c> uppercase form). Those keep matching until they |
| | | 31 | | /// are used or the set is regenerated (FR-2F-20). Crockford's aliases apply to the sixteen |
| | | 32 | | /// only: an older code may really contain an O, an I or an L, and mapping it would make it |
| | | 33 | | /// miss. Any other length matches nothing. |
| | | 34 | | /// </para> |
| | | 35 | | /// <para> |
| | | 36 | | /// <b>Why SHA-256 is enough.</b> A recovery code is not chosen by a person; it is 80 bits |
| | | 37 | | /// drawn uniformly at random, so there is no likely subset for a guesser to start with. |
| | | 38 | | /// Finding one particular code from its digest takes 2^79 SHA-256 evaluations on average, |
| | | 39 | | /// millions of years at the 10^10 a second of a current GPU. The digest is unsalted, so one |
| | | 40 | | /// pass over the space tests every stored code at once, which divides that work by the |
| | | 41 | | /// number of codes stored: at a million codes it is still about 2^60 evaluations — years of |
| | | 42 | | /// one GPU — to recover a single code of some account nobody chose, which is then good only |
| | | 43 | | /// beside that account's password. A slow hash would add little at this entropy and would |
| | | 44 | | /// cost a derivation per unused code on every presentation, since a code is matched against |
| | | 45 | | /// its configuration's whole unused set. A keyed hash, with a pepper held outside the |
| | | 46 | | /// database, is the step to take should the table ever grow by orders of magnitude. The |
| | | 47 | | /// eight-character codes issued before carry fewer bits; regenerating replaces them (UC-40). |
| | | 48 | | /// </para> |
| | | 49 | | /// </remarks> |
| | | 50 | | public static class RecoveryCodes |
| | | 51 | | { |
| | | 52 | | /// <summary>How many codes a set holds (FR-2F-05).</summary> |
| | | 53 | | public const int Count = 10; |
| | | 54 | | |
| | | 55 | | /// <summary>Crockford's base32 alphabet: each character carries five bits.</summary> |
| | | 56 | | public const string Alphabet = "0123456789ABCDEFGHJKMNPQRSTVWXYZ"; |
| | | 57 | | |
| | | 58 | | /// <summary>The characters in a code, hyphens aside.</summary> |
| | | 59 | | public const int Length = 16; |
| | | 60 | | |
| | | 61 | | /// <summary>The characters in a code issued before 80-bit codes, hyphen aside.</summary> |
| | | 62 | | public const int LegacyLength = 8; |
| | | 63 | | |
| | | 64 | | private const int RandomBytes = 10; |
| | | 65 | | private const int GroupLength = 4; |
| | | 66 | | private const int BitsPerCharacter = 5; |
| | | 67 | | |
| | | 68 | | /// <summary>A fresh set of <see cref="Count" /> codes, each with the digest to store for it.</summary> |
| | | 69 | | public static List<IssuedRecoveryCode> Issue() |
| | | 70 | | { |
| | 56 | 71 | | var codes = new List<IssuedRecoveryCode>(Count); |
| | | 72 | | |
| | 1232 | 73 | | for (var i = 0; i < Count; i++) |
| | | 74 | | { |
| | 560 | 75 | | var canonical = NewCanonicalCode(); |
| | | 76 | | |
| | 560 | 77 | | codes.Add(new IssuedRecoveryCode(Display(canonical), Digest(canonical))); |
| | | 78 | | } |
| | | 79 | | |
| | 56 | 80 | | return codes; |
| | | 81 | | } |
| | | 82 | | |
| | | 83 | | /// <summary> |
| | | 84 | | /// The digest <paramref name="presented" /> would be stored under, or <c>null</c> when it |
| | | 85 | | /// cannot be a recovery code of either shape and so matches nothing. See the remarks for |
| | | 86 | | /// what is normalised. |
| | | 87 | | /// </summary> |
| | | 88 | | public static byte[]? DigestOf(string? presented) |
| | | 89 | | { |
| | 164 | 90 | | if (presented is null) |
| | | 91 | | { |
| | 1 | 92 | | return null; |
| | | 93 | | } |
| | | 94 | | |
| | 163 | 95 | | var stripped = new StringBuilder(presented.Length); |
| | | 96 | | |
| | 4701 | 97 | | foreach (var character in presented) |
| | | 98 | | { |
| | 2189 | 99 | | if (character == '-' || char.IsWhiteSpace(character)) |
| | | 100 | | { |
| | | 101 | | continue; |
| | | 102 | | } |
| | | 103 | | |
| | | 104 | | // Neither alphabet has anything outside ASCII, and upper-casing one could turn it into |
| | | 105 | | // a letter that is in them (a dotless i into I, say). |
| | 1857 | 106 | | if (!char.IsAscii(character)) |
| | | 107 | | { |
| | 3 | 108 | | return null; |
| | | 109 | | } |
| | | 110 | | |
| | 1854 | 111 | | stripped.Append(char.ToUpperInvariant(character)); |
| | | 112 | | } |
| | | 113 | | |
| | 160 | 114 | | return stripped.Length switch |
| | 160 | 115 | | { |
| | 78 | 116 | | Length => CurrentDigest(stripped.ToString()), |
| | 58 | 117 | | LegacyLength => LegacyDigest(stripped.ToString()), |
| | 24 | 118 | | _ => null |
| | 160 | 119 | | }; |
| | | 120 | | } |
| | | 121 | | |
| | | 122 | | /// <summary> |
| | | 123 | | /// <paramref name="bytes" /> in <see cref="Alphabet" />, five bits a character, most |
| | | 124 | | /// significant first: the canonical form of a code. Ten bytes give the sixteen characters of |
| | | 125 | | /// <see cref="Length" /> exactly. Public so the mapping can be pinned by known answers; issuing |
| | | 126 | | /// goes through <see cref="Issue" />. |
| | | 127 | | /// </summary> |
| | | 128 | | /// <exception cref="ArgumentException">The length is not a multiple of five bytes.</exception> |
| | | 129 | | public static string Encode(ReadOnlySpan<byte> bytes) |
| | | 130 | | { |
| | 565 | 131 | | if (bytes.Length % 5 != 0) |
| | | 132 | | { |
| | 1 | 133 | | throw new ArgumentException("Base32 without padding needs a multiple of five bytes.", nameof(bytes)); |
| | | 134 | | } |
| | | 135 | | |
| | 564 | 136 | | var code = new StringBuilder(bytes.Length * 8 / BitsPerCharacter); |
| | 564 | 137 | | var buffer = 0; |
| | 564 | 138 | | var bits = 0; |
| | | 139 | | |
| | 12408 | 140 | | foreach (var value in bytes) |
| | | 141 | | { |
| | 5640 | 142 | | buffer = (buffer << 8) | value; |
| | 5640 | 143 | | bits += 8; |
| | | 144 | | |
| | 14664 | 145 | | while (bits >= BitsPerCharacter) |
| | | 146 | | { |
| | 9024 | 147 | | bits -= BitsPerCharacter; |
| | 9024 | 148 | | code.Append(Alphabet[(buffer >> bits) & 0b11111]); |
| | | 149 | | } |
| | | 150 | | } |
| | | 151 | | |
| | 564 | 152 | | return code.ToString(); |
| | | 153 | | } |
| | | 154 | | |
| | | 155 | | /// <summary>Sixteen characters for <see cref="RandomBytes" /> random bytes.</summary> |
| | | 156 | | private static string NewCanonicalCode() |
| | | 157 | | { |
| | 560 | 158 | | Span<byte> random = stackalloc byte[RandomBytes]; |
| | 560 | 159 | | RandomNumberGenerator.Fill(random); |
| | | 160 | | |
| | 560 | 161 | | return Encode(random); |
| | | 162 | | } |
| | | 163 | | |
| | | 164 | | /// <summary><c>XXXX-XXXX-XXXX-XXXX</c>.</summary> |
| | | 165 | | private static string Display(string canonical) => |
| | 560 | 166 | | string.Join('-', Enumerable.Range(0, Length / GroupLength) |
| | 2800 | 167 | | .Select(group => canonical.Substring(group * GroupLength, GroupLength))); |
| | | 168 | | |
| | | 169 | | /// <summary> |
| | | 170 | | /// A sixteen-character code: Crockford's decode aliases applied (O reads as 0, I and L as 1), |
| | | 171 | | /// then hashed in that canonical form. A U, or anything else outside the alphabet, matches |
| | | 172 | | /// nothing. |
| | | 173 | | /// </summary> |
| | | 174 | | private static byte[]? CurrentDigest(string stripped) |
| | | 175 | | { |
| | 78 | 176 | | var canonical = new StringBuilder(Length); |
| | | 177 | | |
| | 2618 | 178 | | foreach (var character in stripped) |
| | | 179 | | { |
| | 1232 | 180 | | var decoded = character switch |
| | 1232 | 181 | | { |
| | 12 | 182 | | 'O' => '0', |
| | 14 | 183 | | 'I' or 'L' => '1', |
| | 1206 | 184 | | _ => character |
| | 1232 | 185 | | }; |
| | | 186 | | |
| | 1232 | 187 | | if (!Alphabet.Contains(decoded)) |
| | | 188 | | { |
| | 2 | 189 | | return null; |
| | | 190 | | } |
| | | 191 | | |
| | 1230 | 192 | | canonical.Append(decoded); |
| | | 193 | | } |
| | | 194 | | |
| | 76 | 195 | | return Digest(canonical.ToString()); |
| | | 196 | | } |
| | | 197 | | |
| | | 198 | | /// <summary> |
| | | 199 | | /// An eight-character code issued before: rebuilt as the exact <c>XXXX-XXXX</c> it was stored |
| | | 200 | | /// as, with no aliases, since its alphabet had every letter. |
| | | 201 | | /// </summary> |
| | | 202 | | private static byte[]? LegacyDigest(string stripped) => |
| | 58 | 203 | | stripped.All(char.IsAsciiLetterOrDigit) |
| | 58 | 204 | | ? Digest($"{stripped[..GroupLength]}-{stripped[GroupLength..]}") |
| | 58 | 205 | | : null; |
| | | 206 | | |
| | 693 | 207 | | private static byte[] Digest(string value) => SHA256.HashData(Encoding.UTF8.GetBytes(value)); |
| | | 208 | | } |