Ted Unangst wrote:
> http://ftp.openbsd.org/pub/OpenBSD/patches/5.9/common/005_crypto.patch.sig
There is an additional chunk in this diff, for s3_pkt.c, that should have not
been included. It adds a memset that will zero a buffer after libssl is done
using it to prevent info leaks. As far as I know, it works fine, but it's not
related to the security advisory. This chunk is only in the 5.9 patch.
If you follow the patch instructions and only rebuild libcrypto, this won't
matter. If you happen to rebuild libssl as well, nothing bad should happen,
but that change didn't go through the normal testing and review process.
We'll work on getting a new patch updated. If you follow the stable CVS
branch, I've already reverted the change. If you apply patches and want to
revert, there's a diff below to apply. Again, if you haven't built libssl, you
don't have to do anything, but I know some people like tidy src trees.
Index: src/lib/libssl/src/ssl/s3_pkt.c
===================================================================
RCS file: /cvs/src/lib/libssl/src/ssl/s3_pkt.c,v
retrieving revision 1.57.2.1
retrieving revision 1.57.2.2
diff -u -p -r1.57.2.1 -r1.57.2.2
--- src/lib/libssl/src/ssl/s3_pkt.c 3 May 2016 12:39:48 -0000 1.57.2.1
+++ src/lib/libssl/src/ssl/s3_pkt.c 4 May 2016 01:10:57 -0000 1.57.2.2
@@ -1,4 +1,4 @@
@@ -956,7 +956,6 @@ start:
memcpy(buf, &(rr->data[rr->off]), n);
if (!peek) {
- memset(&(rr->data[rr->off]), 0, n);
rr->length -= n;
rr->off += n;
if (rr->length == 0) {
