Re: PostgreSQL - Predictable cancel key

[Michael Samuel <mik () miknet net>]

Hi,

On 16 June 2015 at 20:15, Pierre Schweitzer <pierre () reactos org> wrote:

> Well, I've thought about it, but that's still one million seeds and
> the cancel key you're looking for might have been generated after
> several random() call. So, that means perhaps 10 millions values to
> explore? (If we make the hypothesis that for a given initialization,
> it will only draw 10 PRN)

This is CPU, maybe GPU territory. 2^15-2 * 10m =~ 2^38 work offline.


> How relevant would still be your cancel key once you found it?


In the postmaster mainloop:

ConnCreate()
    RandomSalt() -- the md5Salt craziness in the original post
BackendStartup()
    MyCancelKey = PostmasterRandom();

No other calls to random(), srandom() that I can see, and the other calls
to PostmasterRandom() are generating cancel keys for worker procs.

So if you brute the md5Salt you have a pretty good chance of guessing
something that was/is/will be a cancel key of another session.

Regards,
  Michael