Re: bin/52986: netpgpverify broken on sparc64

[Martin Husemann <martin%duskware.de@localhost>]

The following reply was made to PR bin/52986; it has been noted by GNATS.

From: Martin Husemann <martin%duskware.de@localhost>
To: Christos Zoulas <christos%zoulas.com@localhost>
Cc: gnats-bugs%NetBSD.org@localhost, agc%NetBSD.org@localhost, gnats-admin%netbsd.org@localhost,
	netbsd-bugs%netbsd.org@localhost, martin%NetBSD.org@localhost
Subject: Re: bin/52986: netpgpverify broken on sparc64
Date: Mon, 12 Feb 2018 09:34:57 +0100

 On Sun, Feb 11, 2018 at 11:33:28AM -0500, Christos Zoulas wrote:
 > |  But note that this memory is on the stack (or did I miscount the f's?) - I bet
 > |  that is not planned to be so.
 > 
 > it is mmapped so ...
 
 No, in this case it is not (it is verifying a memory hash). I haven't found
 exactly where it is set, but I guess it living on the stack is a bug.
 
 However, there are two more bugs:
 
 Index: bcopy.c
 ===================================================================
 RCS file: /cvsroot/src/common/lib/libc/string/bcopy.c,v
 retrieving revision 1.11
 retrieving revision 1.12
 [..]
 @@ -89,7 +89,7 @@
         unsigned long u;
  
  #if !defined(_KERNEL)
 -       _DIAGASSERT((dst0 && src0) || length == 0);
 +       _DIAGASSERT(length == 0);
  #endif
 
 (inside the memmove implementation used by sparc64) always fires for
 the memmove's called by the BN lib here. This is obviously wrong,
 but: the firing assert does not terminate the program, but during handling
 of it somewhere the stack/memory corruption changing the encoded "hashalg"
 to zero happens.
 
 
 Martin