Re:A Base64 Surprise
by jedisct1 on Saturday January 29, 2022 @05:07PM
(#30131046)
The choice to ignore or reject non-null trailing padding bits is really implementation dependent.
Python and Zig reject invalid padding.
PHP's base64_decode() function accepts it, but the sodium_base642bin() function doesn't.
Since Rust doesn't have base64 support, it depends on the crate being used. The ct-codecs crate rejects invalid padding, the base64 crate doesn't.
So... it depends.
But that issue is well known, and I think most modern implementations reject invalid padding. And in the context of cryptographic operations, you are supposed to use codecs from the cryptographic libraries you are using anyway. These implementations avoid side channels, and, of course, reject non-canonical encodings.
Re:A Base64 Surprise
by cendyne on Saturday January 29, 2022 @05:52PM
(#30131426)
Your example [1] on twitter for PHP is actually quite telling.
I do not think it is well known among application developers, which is why I wrote this post.
In fact, my base64 implementation elsewhere is vulnerable to this and it never occurred to me.
Thank you for the rust library recommendation, I was using base64ct.
Re:A Base64 Surprise
by Koshkin on Saturday January 29, 2022 @06:29PM
(#30131717)
Interestingly (and somewhat confusingly), the number of the padding characters appears to indicate both how many (6-bit) characters are missing from the last quadruple in the output of the encoder and, at the same time, how many (8-bit) bytes will be trimmed from the last triple by the decoder.
Re:A Base64 Surprise
by ncmncm on Saturday January 29, 2022 @06:18PM
(#30131627)
This has nothing to do with Rust, as suggested in pp#1, and all the guff leading up to the real topic seems to be an attempt at some sort of perverted virtue signaling.
Python and Zig reject invalid padding.
PHP's base64_decode() function accepts it, but the sodium_base642bin() function doesn't.
Since Rust doesn't have base64 support, it depends on the crate being used. The ct-codecs crate rejects invalid padding, the base64 crate doesn't.
So... it depends.
But that issue is well known, and I think most modern implementations reject invalid padding. And in the context of cryptographic operations, you are supposed to use codecs from the cryptographic libraries you are using anyway. These implementations avoid side channels, and, of course, reject non-canonical encodings.
I do not think it is well known among application developers, which is why I wrote this post. In fact, my base64 implementation elsewhere is vulnerable to this and it never occurred to me.
Thank you for the rust library recommendation, I was using base64ct.
[1]: https://archive.is/RZwlH
Other than that, cool article :)