Deprecated in favor of modules requiring compilation; is there no good solution using node's built-in crypto module?

[boutell]

I understand you've deprecated this module. I'm writing to ask if there are alternatives you consider secure that take advantage of node's built-in crypto module as a backend and do not require the compilation of custom C++ code beyond that. That could reduce the overall amount of attackable C++ code in the system. It's also convenient where compilation is unwelcome, although I recognize that's a secondary concern.

I'm also wondering if there's any urgency to migrate off this module completely and force users to generate new passwords. I've been using it with its default settings for some time.

[davidwood]

Both bcrypt and scrypt are password-based key derivation functions. The node crypto module HAS an equivalent style function, PBKDF2, which has been around for a while (I recall it being in 0.6.x if not earlier). The issue from a compatibility standpoint is that password-hash exposes as synchronous API, whereas PBKDF2 has been asynchronous until 0.10.0, when they added a synchronous version.

Let me think about the best way to add support for this. Ideally, generate and verify become async, but perhaps they supports both sync and async depending on whether or not a callback is passed.

Out of curiosity, what makes you not trust C++ code that has the attention of security researchers? Do you consider that less secure than a JavaScript module?

[boutell]

I separately queried the bcrypt devs, who gave an excellent explanation of
why it is trustworthy.

I occasionally have issues with self-hosted customers who are stubborn
about allowing the compiler to be on the system at all, which can make
installing these modules a pain, but that shouldn't drive the community's
security decisions.

On Tue, Aug 5, 2014 at 9:07 AM, David Wood notifications@github.com wrote:

Both bcrypt and scrypt are password-based key derivation functions. The
node crypto module HAS an equivalent style function, PBKDF2, which has been
around for a while (I recall it being in 0.6.x if not earlier). The issue
from a compatibility standpoint is that password-hash exposes as
synchronous API, whereas PBKDF2 has been asynchronous until 0.10.0, when
they added a synchronous version.

Let me think about the best way to add support for this. Ideally, generate
and verify become async, but perhaps they supports both sync and async
depending on whether or not a callback is passed.

Out of curiosity, what makes you not trust C++ code that has the attention
of security researchers? Do you consider that less secure than a JavaScript
module?

—
Reply to this email directly or view it on GitHub
#10 (comment)
.

*THOMAS BOUTELL, *DEV & OPS
P'UNK AVENUE | (215) 755-1330 | punkave.com

[davidwood]

PBKDF2, bcrypt, and scrypt are similar in general approach, but the general consensus is that scrypt > bcrypt > PBKDF2. That being said PBKDF2 has received more security professional review and is what NIST recommends.

I'll work PBKDF2 into the module this week.