summaryrefslogtreecommitdiff
path: root/base32/src/lib.rs
diff options
context:
space:
mode:
authorRobin Krahl <robin.krahl@ireas.org>2019-01-06 00:20:04 +0100
committerDaniel Mueller <deso@posteo.net>2019-01-13 18:37:08 -0800
commitcb5b22d0886a0a8c19cf1afeef55a8245357cf20 (patch)
treecfcc555186317bac4f4383a9380e0d6ef6d58459 /base32/src/lib.rs
parent38e7e1d92ce4aea7840d78bcda6246bd61c423ac (diff)
downloadnitrocli-cb5b22d0886a0a8c19cf1afeef55a8245357cf20.tar.gz
nitrocli-cb5b22d0886a0a8c19cf1afeef55a8245357cf20.tar.bz2
Add the base32 crate in version 0.4.0 as a dependency
To parse OTP secrets in base32 representation, we need a new dependency: the base32 crate. Import subrepo base32/:base32 at a74cd9246fc0e08d6f5cfcb644bfdf76dd438613
Diffstat (limited to 'base32/src/lib.rs')
-rw-r--r--base32/src/lib.rs221
1 files changed, 221 insertions, 0 deletions
diff --git a/base32/src/lib.rs b/base32/src/lib.rs
new file mode 100644
index 0000000..e21cf97
--- /dev/null
+++ b/base32/src/lib.rs
@@ -0,0 +1,221 @@
+#[cfg(test)]
+extern crate quickcheck;
+#[cfg(test)]
+extern crate rand;
+
+use std::cmp::min;
+
+#[derive(Copy,Clone)]
+pub enum Alphabet {
+ RFC4648 { padding: bool },
+ Crockford,
+}
+
+const RFC4648_ALPHABET: &'static [u8] = b"ABCDEFGHIJKLMNOPQRSTUVWXYZ234567";
+const CROCKFORD_ALPHABET: &'static [u8] = b"0123456789ABCDEFGHJKMNPQRSTVWXYZ";
+
+pub fn encode(alphabet: Alphabet, data: &[u8]) -> String {
+ let (alphabet, padding) = match alphabet {
+ Alphabet::RFC4648 { padding } => (RFC4648_ALPHABET, padding),
+ Alphabet::Crockford => (CROCKFORD_ALPHABET, false),
+ };
+ let mut ret = Vec::with_capacity((data.len()+3)/4*5);
+
+ for chunk in data.chunks(5) {
+ let buf = {
+ let mut buf = [0u8; 5];
+ for (i, &b) in chunk.iter().enumerate() {
+ buf[i] = b;
+ }
+ buf
+ };
+ ret.push(alphabet[((buf[0] & 0xF8) >> 3) as usize]);
+ ret.push(alphabet[(((buf[0] & 0x07) << 2) | ((buf[1] & 0xC0) >> 6)) as usize]);
+ ret.push(alphabet[((buf[1] & 0x3E) >> 1) as usize]);
+ ret.push(alphabet[(((buf[1] & 0x01) << 4) | ((buf[2] & 0xF0) >> 4)) as usize]);
+ ret.push(alphabet[(((buf[2] & 0x0F) << 1) | (buf[3] >> 7)) as usize]);
+ ret.push(alphabet[((buf[3] & 0x7C) >> 2) as usize]);
+ ret.push(alphabet[(((buf[3] & 0x03) << 3) | ((buf[4] & 0xE0) >> 5)) as usize]);
+ ret.push(alphabet[(buf[4] & 0x1F) as usize]);
+ }
+
+ if data.len() % 5 != 0 {
+ let len = ret.len();
+ let num_extra = 8-(data.len()%5*8+4)/5;
+ if padding {
+ for i in 1..num_extra+1 {
+ ret[len-i] = b'=';
+ }
+ } else {
+ ret.truncate(len-num_extra);
+ }
+ }
+
+ String::from_utf8(ret).unwrap()
+}
+
+const RFC4648_INV_ALPHABET: [i8; 43] = [-1, -1, 26, 27, 28, 29, 30, 31, -1, -1, -1, -1, -1, 0, -1, -1, -1, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25];
+const CROCKFORD_INV_ALPHABET: [i8; 43] = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, -1, -1, -1, -1, -1, -1, -1, 10, 11, 12, 13, 14, 15, 16, 17, 1, 18, 19, 1, 20, 21, 0, 22, 23, 24, 25, 26, -1, 27, 28, 29, 30, 31];
+
+pub fn decode(alphabet: Alphabet, data: &str) -> Option<Vec<u8>> {
+ if !data.is_ascii() {
+ return None;
+ }
+ let data = data.as_bytes();
+ let alphabet = match alphabet {
+ Alphabet::RFC4648 {..} => RFC4648_INV_ALPHABET,
+ Alphabet::Crockford => CROCKFORD_INV_ALPHABET
+ };
+ let mut unpadded_data_length = data.len();
+ for i in 1..min(6, data.len())+1 {
+ if data[data.len() - i] != b'=' {
+ break;
+ }
+ unpadded_data_length -= 1;
+ }
+ let output_length = unpadded_data_length*5/8;
+ let mut ret = Vec::with_capacity((output_length+4)/5*5);
+ for chunk in data.chunks(8) {
+ let buf = {
+ let mut buf = [0u8; 8];
+ for (i, &c) in chunk.iter().enumerate() {
+ match alphabet.get(c.to_ascii_uppercase().wrapping_sub(b'0') as usize) {
+ Some(&-1) | None => return None,
+ Some(&value) => buf[i] = value as u8,
+ };
+ }
+ buf
+ };
+ ret.push((buf[0] << 3) | (buf[1] >> 2));
+ ret.push((buf[1] << 6) | (buf[2] << 1) | (buf[3] >> 4));
+ ret.push((buf[3] << 4) | (buf[4] >> 1));
+ ret.push((buf[4] << 7) | (buf[5] << 2) | (buf[6] >> 3));
+ ret.push((buf[6] << 5) | buf[7]);
+ }
+ ret.truncate(output_length);
+ Some(ret)
+}
+
+#[cfg(test)]
+#[allow(dead_code, unused_attributes)]
+mod test {
+ use super::{encode, decode};
+ use super::Alphabet::{Crockford, RFC4648};
+ use quickcheck;
+ use std;
+ use rand::Rng;
+
+ #[derive(Clone)]
+ struct B32 {
+ c: u8
+ }
+
+ impl quickcheck::Arbitrary for B32 {
+ fn arbitrary<G: quickcheck::Gen>(g: &mut G) -> B32 {
+ let alphabet = b"0123456789ABCDEFGHJKMNPQRSTVWXYZ";
+ B32 {
+ c: alphabet[g.gen_range(0, alphabet.len())]
+ }
+ }
+ }
+
+ impl std::fmt::Debug for B32 {
+ fn fmt(&self, f: &mut std::fmt::Formatter) -> Result<(), std::fmt::Error> {
+ (self.c as char).fmt(f)
+ }
+ }
+
+ #[test]
+ fn masks_crockford() {
+ assert_eq!(encode(Crockford, &[0xF8, 0x3E, 0x0F, 0x83, 0xE0]), "Z0Z0Z0Z0");
+ assert_eq!(encode(Crockford, &[0x07, 0xC1, 0xF0, 0x7C, 0x1F]), "0Z0Z0Z0Z");
+ assert_eq!(decode(Crockford, "Z0Z0Z0Z0").unwrap(), [0xF8, 0x3E, 0x0F, 0x83, 0xE0]);
+ assert_eq!(decode(Crockford, "0Z0Z0Z0Z").unwrap(), [0x07, 0xC1, 0xF0, 0x7C, 0x1F]);
+ }
+
+ #[test]
+ fn masks_rfc4648() {
+ assert_eq!(encode(RFC4648 { padding: true }, &[0xF8, 0x3E, 0x7F, 0x83, 0xE7]), "7A7H7A7H");
+ assert_eq!(encode(RFC4648 { padding: true }, &[0x77, 0xC1, 0xF7, 0x7C, 0x1F]), "O7A7O7A7");
+ assert_eq!(decode(RFC4648 { padding: true }, "7A7H7A7H").unwrap(), [0xF8, 0x3E, 0x7F, 0x83, 0xE7]);
+ assert_eq!(decode(RFC4648 { padding: true }, "O7A7O7A7").unwrap(), [0x77, 0xC1, 0xF7, 0x7C, 0x1F]);
+ assert_eq!(encode(RFC4648 { padding: true }, &[0xF8, 0x3E, 0x7F, 0x83]), "7A7H7AY=");
+ }
+
+ #[test]
+ fn masks_unpadded_rfc4648() {
+ assert_eq!(encode(RFC4648 { padding: false }, &[0xF8, 0x3E, 0x7F, 0x83, 0xE7]), "7A7H7A7H");
+ assert_eq!(encode(RFC4648 { padding: false }, &[0x77, 0xC1, 0xF7, 0x7C, 0x1F]), "O7A7O7A7");
+ assert_eq!(decode(RFC4648 { padding: false }, "7A7H7A7H").unwrap(), [0xF8, 0x3E, 0x7F, 0x83, 0xE7]);
+ assert_eq!(decode(RFC4648 { padding: false }, "O7A7O7A7").unwrap(), [0x77, 0xC1, 0xF7, 0x7C, 0x1F]);
+ assert_eq!(encode(RFC4648 { padding: false }, &[0xF8, 0x3E, 0x7F, 0x83]), "7A7H7AY");
+ }
+
+ #[test]
+ fn padding() {
+ let num_padding = [0, 6, 4, 3, 1];
+ for i in 1..6 {
+ let encoded = encode(RFC4648 { padding: true }, (0..(i as u8)).collect::<Vec<u8>>().as_ref());
+ assert_eq!(encoded.len(), 8);
+ for j in 0..(num_padding[i % 5]) {
+ assert_eq!(encoded.as_bytes()[encoded.len()-j-1], b'=');
+ }
+ for j in 0..(8 - num_padding[i % 5]) {
+ assert!(encoded.as_bytes()[j] != b'=');
+ }
+ }
+ }
+
+ #[test]
+ fn invertible_crockford() {
+ fn test(data: Vec<u8>) -> bool {
+ decode(Crockford, encode(Crockford, data.as_ref()).as_ref()).unwrap() == data
+ }
+ quickcheck::quickcheck(test as fn(Vec<u8>) -> bool)
+ }
+
+ #[test]
+ fn invertible_rfc4648() {
+ fn test(data: Vec<u8>) -> bool {
+ decode(RFC4648 { padding: true }, encode(RFC4648 { padding: true }, data.as_ref()).as_ref()).unwrap() == data
+ }
+ quickcheck::quickcheck(test as fn(Vec<u8>) -> bool)
+ }
+ #[test]
+ fn invertible_unpadded_rfc4648() {
+ fn test(data: Vec<u8>) -> bool {
+ decode(RFC4648 { padding: false }, encode(RFC4648 { padding: false }, data.as_ref()).as_ref()).unwrap() == data
+ }
+ quickcheck::quickcheck(test as fn(Vec<u8>) -> bool)
+ }
+
+ #[test]
+ fn lower_case() {
+ fn test(data: Vec<B32>) -> bool {
+ let data: String = data.iter().map(|e| e.c as char).collect();
+ decode(Crockford, data.as_ref()) == decode(Crockford, data.to_ascii_lowercase().as_ref())
+ }
+ quickcheck::quickcheck(test as fn(Vec<B32>) -> bool)
+ }
+
+ #[test]
+ #[allow(non_snake_case)]
+ fn iIlL1_oO0() {
+ assert_eq!(decode(Crockford, "IiLlOo"), decode(Crockford, "111100"));
+ }
+
+ #[test]
+ fn invalid_chars_crockford() {
+ assert_eq!(decode(Crockford, ","), None)
+ }
+
+ #[test]
+ fn invalid_chars_rfc4648() {
+ assert_eq!(decode(RFC4648 { padding: true }, ","), None)
+ }
+
+ #[test]
+ fn invalid_chars_unpadded_rfc4648() {
+ assert_eq!(decode(RFC4648 { padding: false }, ","), None)
+ }
+}