Re: glibc crypt(3), crypt_r(3), PHP crypt() may use alloca()

[Steve Grubb <sgrubb () redhat com>]

On Wednesday, November 16, 2011 09:22:17 PM Solar Designer wrote:
> On Tue, Nov 15, 2011 at 06:13:24AM +0400, Solar Designer wrote:
> > Alternatively, crypt(3) and crypt_r(3) (and the reference code for
> > SHA-crypt?) could refuse to work on overly long key or/and salt strings,
> > but then the question is what they should do on error.
>
> Here's another related option:
>
>       if (strlen(key) > 100000 || strlen(salt) > 100000)
>               abort();
>
> (or something like this).  Ridiculous?  Sure, but it's better than
> overwriting another thread's stack or the heap with somewhat higher
> lengths, and 100001 chars is not a more reasonable password length to
> support than, say, 2 million or 10 million (typical thread stack sizes).
>
> So if we can't decide on a proper fix (does anyone besides me even
> care?), something as trivial as the above would be an improvement.

raise(SIGKILL) might be better because abort requests a core dump and you are in 
crypto code.

-Steve