support decoding byte slices

All of the backend infrastructure was in place to support &[u8], and some functions already take `impl AsRef<[u8]>`. However the main decode() and decode_header() functions require a &str. This change updates a few internal signatures and adds a _bytes() version of decode and decode_header. When you're doing many requests per second, the cost of doing an extra utf-8 check over header payloads is significant. By supporting a &[u8] decode, users can let base64 and the crypto implementation in question handle its own bytewise validity. They already do this today in addition to the extra utf-8 scan.
Keats · Apr 21, 2024 · e340d75 · e340d75
1 parent afbb44e
commit e340d75
Show file tree

Hide file tree

Showing 7 changed files with 102 additions and 20 deletions.
diff --git a/benches/jwt.rs b/benches/jwt.rs
@@ -1,5 +1,8 @@
 use criterion::{black_box, criterion_group, criterion_main, Criterion};
-use jsonwebtoken::{decode, encode, Algorithm, DecodingKey, EncodingKey, Header, Validation};
+use jsonwebtoken::{
+    decode, decode_bytes, decode_header, decode_header_bytes, encode, Algorithm, DecodingKey,
+    EncodingKey, Header, Validation,
+};
 use serde::{Deserialize, Serialize};
 
 #[derive(Debug, PartialEq, Clone, Serialize, Deserialize)]
@@ -18,18 +21,47 @@ fn bench_encode(c: &mut Criterion) {
 }
 
 fn bench_decode(c: &mut Criterion) {
-    let token = "eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJzdWIiOiIxMjM0NTY3ODkwIiwibmFtZSI6IkpvaG4gRG9lIiwiYWRtaW4iOnRydWV9.TJVA95OrM7E2cBab30RMHrHDcEfxjoYZgeFONFh7HgQ";
+    let token = b"eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJzdWIiOiIxMjM0NTY3ODkwIiwibmFtZSI6IkpvaG4gRG9lIiwiYWRtaW4iOnRydWV9.TJVA95OrM7E2cBab30RMHrHDcEfxjoYZgeFONFh7HgQ";
     let key = DecodingKey::from_secret("secret".as_ref());
 
-    c.bench_function("bench_decode", |b| {
+    let mut group = c.benchmark_group("decode");
+    group.throughput(criterion::Throughput::Bytes(token.len() as u64));
+
+    group.bench_function("bytes", |b| {
         b.iter(|| {
-            decode::<Claims>(
+            decode_bytes::<Claims>(
                 black_box(token),
                 black_box(&key),
                 black_box(&Validation::new(Algorithm::HS256)),
             )
         })
     });
+
+    group.bench_function("str", |b| {
+        b.iter(|| {
+            decode::<Claims>(
+                // Simulate the cost of validating &str before decoding
+                black_box(std::str::from_utf8(black_box(token)).expect("valid utf8")),
+                black_box(&key),
+                black_box(&Validation::new(Algorithm::HS256)),
+            )
+        })
+    });
+
+    drop(group);
+    let mut group = c.benchmark_group("header");
+    group.throughput(criterion::Throughput::Bytes(token.len() as u64));
+
+    group.bench_function("str", |b| {
+        b.iter(|| {
+            decode_header(
+                // Simulate the cost of validating &str before decoding
+                black_box(std::str::from_utf8(black_box(token)).expect("valid utf8")),
+            )
+        })
+    });
+
+    group.bench_function("bytes", |b| b.iter(|| decode_header_bytes(black_box(token))));
 }
 
 criterion_group!(benches, bench_encode, bench_decode);

diff --git a/src/crypto/mod.rs b/src/crypto/mod.rs
@@ -46,7 +46,7 @@ pub fn sign(message: &[u8], key: &EncodingKey, algorithm: Algorithm) -> Result<S
 /// See Ring docs for more details
 fn verify_ring(
     alg: &'static dyn signature::VerificationAlgorithm,
-    signature: &str,
+    signature: impl AsRef<[u8]>,
     message: &[u8],
     key: &[u8],
 ) -> Result<bool> {
@@ -66,16 +66,17 @@ fn verify_ring(
 ///
 /// `message` is base64(header) + "." + base64(claims)
 pub fn verify(
-    signature: &str,
+    signature: impl AsRef<[u8]>,
     message: &[u8],
     key: &DecodingKey,
     algorithm: Algorithm,
 ) -> Result<bool> {
+    let signature = signature.as_ref();
     match algorithm {
         Algorithm::HS256 | Algorithm::HS384 | Algorithm::HS512 => {
             // we just re-sign the message with the key and compare if they are equal
             let signed = sign(message, &EncodingKey::from_secret(key.as_bytes()), algorithm)?;
-            Ok(verify_slices_are_equal(signature.as_ref(), signed.as_ref()).is_ok())
+            Ok(verify_slices_are_equal(signature, signed.as_ref()).is_ok())
         }
         Algorithm::ES256 | Algorithm::ES384 => verify_ring(
             ecdsa::alg_to_ec_verification(algorithm),

diff --git a/src/crypto/rsa.rs b/src/crypto/rsa.rs
@@ -51,7 +51,7 @@ pub(crate) fn sign(
 /// Checks that a signature is valid based on the (n, e) RSA pubkey components
 pub(crate) fn verify_from_components(
     alg: &'static signature::RsaParameters,
-    signature: &str,
+    signature: impl AsRef<[u8]>,
     message: &[u8],
     components: (&[u8], &[u8]),
 ) -> Result<bool> {

diff --git a/src/decoding.rs b/src/decoding.rs
@@ -204,11 +204,11 @@ impl DecodingKey {
 /// Verify signature of a JWT, and return header object and raw payload
 ///
 /// If the token or its signature is invalid, it will return an error.
-fn verify_signature<'a>(
-    token: &'a str,
+fn verify_signature_bytes<'a>(
+    token: &'a [u8],
     key: &DecodingKey,
     validation: &Validation,
-) -> Result<(Header, &'a str)> {
+) -> Result<(Header, &'a [u8])> {
     if validation.validate_signature && validation.algorithms.is_empty() {
         return Err(new_error(ErrorKind::MissingAlgorithm));
     }
@@ -221,15 +221,15 @@ fn verify_signature<'a>(
         }
     }
 
-    let (signature, message) = expect_two!(token.rsplitn(2, '.'));
-    let (payload, header) = expect_two!(message.rsplitn(2, '.'));
+    let (signature, message) = expect_two!(token.rsplitn(2, |b| *b == b'.'));
+    let (header, payload) = expect_two!(message.splitn(2, |b| *b == b'.'));
     let header = Header::from_encoded(header)?;
 
     if validation.validate_signature && !validation.algorithms.contains(&header.alg) {
         return Err(new_error(ErrorKind::InvalidAlgorithm));
     }
 
-    if validation.validate_signature && !verify(signature, message.as_bytes(), key, header.alg)? {
+    if validation.validate_signature && !verify(signature, message, key, header.alg)? {
         return Err(new_error(ErrorKind::InvalidSignature));
     }
 
@@ -259,7 +259,38 @@ pub fn decode<T: DeserializeOwned>(
     key: &DecodingKey,
     validation: &Validation,
 ) -> Result<TokenData<T>> {
-    match verify_signature(token, key, validation) {
+    decode_bytes(token.as_bytes(), key, validation)
+}
+
+/// Decode and validate a JWT
+///
+/// If the token or its signature is invalid or the claims fail validation, it will return an error.
+///
+/// This differs from decode() in the case that you only have bytes. By decoding as bytes you can
+/// avoid taking a pass over your bytes to validate them as a utf-8 string. Since the decoding and
+/// validation is all done in terms of bytes, the &str step is unnecessary.
+/// If you already have a &str, decode is more convenient. If you have bytes, consider using this.
+///
+/// ```rust
+/// use serde::{Deserialize, Serialize};
+/// use jsonwebtoken::{decode_bytes, DecodingKey, Validation, Algorithm};
+///
+/// #[derive(Debug, Serialize, Deserialize)]
+/// struct Claims {
+///    sub: String,
+///    company: String
+/// }
+///
+/// let token = b"a.jwt.token";
+/// // Claims is a struct that implements Deserialize
+/// let token_message = decode_bytes::<Claims>(token, &DecodingKey::from_secret("secret".as_ref()), &Validation::new(Algorithm::HS256));
+/// ```
+pub fn decode_bytes<T: DeserializeOwned>(
+    token: &[u8],
+    key: &DecodingKey,
+    validation: &Validation,
+) -> Result<TokenData<T>> {
+    match verify_signature_bytes(token, key, validation) {
         Err(e) => Err(e),
         Ok((header, claims)) => {
             let decoded_claims = DecodedJwtPartClaims::from_jwt_part_claims(claims)?;
@@ -286,3 +317,19 @@ pub fn decode_header(token: &str) -> Result<Header> {
     let (_, header) = expect_two!(message.rsplitn(2, '.'));
     Header::from_encoded(header)
 }
+
+/// Decode a JWT without any signature verification/validations and return its [Header](struct.Header.html).
+///
+/// If the token has an invalid format (ie 3 parts separated by a `.`), it will return an error.
+///
+/// ```rust
+/// use jsonwebtoken::decode_header_bytes;
+///
+/// let token = b"a.jwt.token";
+/// let header = decode_header_bytes(token);
+/// ```
+pub fn decode_header_bytes(token: &[u8]) -> Result<Header> {
+    let (_, message) = expect_two!(token.rsplitn(2, |b| *b == b'.'));
+    let (_, header) = expect_two!(message.rsplitn(2, |b| *b == b'.'));
+    Header::from_encoded(header)
+}
diff --git a/src/lib.rs b/src/lib.rs
@@ -18,7 +18,9 @@ mod serialization;
 mod validation;
 
 pub use algorithms::Algorithm;
-pub use decoding::{decode, decode_header, DecodingKey, TokenData};
+pub use decoding::{
+    decode, decode_bytes, decode_header, decode_header_bytes, DecodingKey, TokenData,
+};
 pub use encoding::{encode, EncodingKey};
 pub use header::Header;
 pub use validation::{get_current_timestamp, Validation};
diff --git a/tests/ecdsa/mod.rs b/tests/ecdsa/mod.rs
@@ -26,7 +26,7 @@ fn round_trip_sign_verification_pk8() {
     let encrypted =
         sign(b"hello world", &EncodingKey::from_ec_der(privkey), Algorithm::ES256).unwrap();
     let is_valid =
-        verify(&encrypted, b"hello world", &DecodingKey::from_ec_der(pubkey), Algorithm::ES256)
+        verify(encrypted, b"hello world", &DecodingKey::from_ec_der(pubkey), Algorithm::ES256)
             .unwrap();
     assert!(is_valid);
 }
@@ -41,7 +41,7 @@ fn round_trip_sign_verification_pem() {
         sign(b"hello world", &EncodingKey::from_ec_pem(privkey_pem).unwrap(), Algorithm::ES256)
             .unwrap();
     let is_valid = verify(
-        &encrypted,
+        encrypted,
         b"hello world",
         &DecodingKey::from_ec_pem(pubkey_pem).unwrap(),
         Algorithm::ES256,

diff --git a/tests/eddsa/mod.rs b/tests/eddsa/mod.rs
@@ -26,7 +26,7 @@ fn round_trip_sign_verification_pk8() {
     let encrypted =
         sign(b"hello world", &EncodingKey::from_ed_der(privkey), Algorithm::EdDSA).unwrap();
     let is_valid =
-        verify(&encrypted, b"hello world", &DecodingKey::from_ed_der(pubkey), Algorithm::EdDSA)
+        verify(encrypted, b"hello world", &DecodingKey::from_ed_der(pubkey), Algorithm::EdDSA)
             .unwrap();
     assert!(is_valid);
 }
@@ -41,7 +41,7 @@ fn round_trip_sign_verification_pem() {
         sign(b"hello world", &EncodingKey::from_ed_pem(privkey_pem).unwrap(), Algorithm::EdDSA)
             .unwrap();
     let is_valid = verify(
-        &encrypted,
+        encrypted,
         b"hello world",
         &DecodingKey::from_ed_pem(pubkey_pem).unwrap(),
         Algorithm::EdDSA,