Tons of "Failed to decrypt and verify packet"

[Julien Muchembled]

Le 04/07/14 01:50, Pedro Côrte-Real a écrit :
> On Sun, Apr 6, 2014 at 8:58 PM, Julien Muchembled <jm at jmuchemb.eu> wrote:
>> I tried to bisect. Last good revision is
>>> e42bd60 Fix typos in the documentation.
>> And I get errors with
>>> c1703ea Remove an unused variable.
> 
> Sorry to chime in the middle. Maybe a clue. This sounds like something
> is writing beyond bounds in the stack. Removing an unused variable on
> the stack would change the part of the stack that is "safe" to
> overwrite to.

So I continued to bisect.
Commit 0da0728 ("Use AES-256-GCM for the SPTPS protocol") is really the first bad commit.

> Maybe Guus and Julien are not testing on the same architecture and
> thus the stack doesn't grow the same way and doesn't show the same
> symptoms. I had the same issue when coding with a friend that was
> using a PPC mac a few years ago. Something that was a crashing for me
> was only a memory corruption for him or vice versa.

I'm testing on x86_64, which is the most common architecture today.

> Looking at the code I see:
> 
> fd = accept(sock, &sa.sa, &len);
> 
> where len is actually sizeof(sa) and not sizeof(sa.sa). Is that correct?

I'd say it does not matter. But probably less confusing to change sizeof(sa) into sizeof(sa.sa)