Cert validation behaviour change in 2.2.0

[will118]

I've noticed a change in behaviour with the following:

fn main() {
    let cert = "-----BEGIN CERTIFICATE-----\nINVALID\n-----END CERTIFICATE-----";
    let cert = rustls_pemfile::certs(&mut cert.as_bytes())
        .next()
        .unwrap()
        .unwrap();
    dbg!(cert);
}

[package]
name = "bad-cert"
version = "0.1.0"
edition = "2021"

[dependencies]
rustls-pemfile = "=2.1.2"
# rustls-pemfile = "=2.2.0"

In 2.1.2 I see:

Custom { kind: InvalidData, error: "InvalidLastSymbol(6, 68)" }

but in 2.2.0 I get:

[src/main.rs:7:5] cert = CertificateDer(
    0x20d5402c80,
)

I'm not confident what the docs mean by:

Filters out any PEM sections that are not certificates and yields errors if a problem occurs while trying to extract a certificate.

...but regardless the change surprised me and was caught by a test.

Is it expected?

[ctz]

Thanks for reporting this!

INVALID is actually valid base64, ignoring padding. The base64 crate has a behaviour where it requires the last codepoint has unused bits that are cleared (this is important for users where encodings must be canonical, but I don't think PEM is one of these -- see https://datatracker.ietf.org/doc/html/rfc4648#section-3.5 for a description). Whereas, other implementations don't do that:

$ cat huh.go
package main

import (
	"encoding/base64"
	"fmt"
)

func main() {
	decoded, err := base64.StdEncoding.WithPadding(base64.NoPadding).DecodeString("INVALID")
	if err != nil {
		fmt.Println("decode error:", err)
		return
	}
	fmt.Printf("%x\n", string(decoded))
}

$ go run huh.go 
20d5402c80

I would recommend using a more overtly wrong base64 encoding in your test, such as * which will not be accepted by any implementation.

edit: altered to language to remove implication that base64 crate does something non-standard; it is just optional according to RFC4648.

[djc]

(Sorry, read your issue report wrong.)

[will118]

OK, thank you for the explanation.

I think there was a misunderstanding (on our part) about the validation performed by certs(..) (valid PEM vs valid "certificate").

Now I've taken the time to understand, I don't think the change in behaviour is that big a deal...

I'm happy to close this.