[NFR] Phalcon\Security to support $2y$ bcrypt hashes

[viktoras25]

Currently only $2a$ bcrypt hash is available. An option to choose $2y$ would be very nice, so that it is compatible with PHP 5.5+ function password_hash();

[ghost]

$2a$ and $2y$ generate the same hashes as far as I can tell

[ghost]

@phalcon Basing on http://www.php.net/security/crypt_blowfish.php maybe we should drop official support for PHP up to 5.3.7 exclusively? PHP up to 5.3.10 have a lot of security vulnerabilities and I doubt that they are used in production.

[viktoras25]

I'm not an expert, but as far as I understand according to this:
http://blog.ircmaxell.com/2012/12/seven-ways-to-screw-up-bcrypt.html
$2y$ is preferable. And it is obvious that $2a$ is much, much faster to hash (according to my tests password_hash/verify, that uses $2y$ is 500-1200 times slower), which also means faster to bruteforce.

[ghost]

And it is obvious that $2a$ is much, much faster to hash

Interesting because the code is absolutely the same :-)

My tests:

$ time php -r 'for ($i=0; $i<1000000; ++$i) crypt("test", "$2a$0123456789012345678912");'

real    0m5.469s
user    0m5.450s
sys     0m0.024s

$ time php -r 'for ($i=0; $i<1000000; ++$i) crypt("test", "$2x$0123456789012345678912");'

real    0m5.452s
user    0m5.450s
sys     0m0.008s

$ time php -r 'for ($i=0; $i<1000000; ++$i) crypt("test", "$2y$0123456789012345678912");'

real    0m5.511s
user    0m5.475s
sys     0m0.016s

http://lxr.php.net/xref/PHP_5_5/ext/standard/crypt_blowfish.c#548

The bug affected the possibility of collisions, not the number of iterations so it is expected that the has generation time will be nearly the same.

[ghost]

If you mean that password_hash() is slower than crypt(), then yes, if you don't provide a salt to password_hash() it will generate one from /dev/urandom (link)

Other than that, both functions use php_crypt() API internally.

[viktoras25]

I have no other explanation other than there was an error in my tests.

There is still difference about 4 times between Phalcon\Security and password_hash generated hashes, but I presume this is due to different ways of salt generating.

[ghost]

You closed this too early :-) I was adding support for additional hash types :-)

[viktoras25]

Oh, sorry :)

[ghost]

Fix submitted.

See https://github.com/sjinks/cphalcon/blob/issue-1971/ext/tests/issue-1971.phpt for usage example

[viktoras25]

Impressive!