The version 3.5.3 triggers undefined behaviour with the usan sanatizer.
Updating LibreSSSL to 3.7.0 does unfortunately does not fix the issue but
at least we are now using a current version.
Signed-off-by: Arne Schwabe
---
.github/workflows/build.yaml | 4 ++--
1 file changed, 2 insertions(+), 2
The -fno-sanitize-recover=all flag ensures that for all errors we actually
abort the tests in the automated testing and not just print some errors in red
that nobody sess. Also add the undefined tests to catch more bugs.
For libreSSL we do not add the udefined behaviour as we have (even with the
l
The undefined behaviour USAN clang checker found this. The optimiser
of clang/gcc will optimise the memcpy away in the auth_token case and output
excactly the same assembly on amd64/arm64 but it is still better to not rely
on undefined behaviour.
Signed-off-by: Arne Schwabe
---
src/openvpn/auth_
The undefined behaviour USAN clang checker found this.
This fix is a bit messy but so are the original structures.
Signed-off-by: Arne Schwabe
---
src/openvpn/route.c | 32 +++-
1 file changed, 23 insertions(+), 9 deletions(-)
diff --git a/src/openvpn/route.c b/src/
This allows identifying the source of undefined behaviour more easily
from the github action logs.
Signed-off-by: Arne Schwabe
---
.github/workflows/build.yaml | 5 +
1 file changed, 5 insertions(+)
diff --git a/.github/workflows/build.yaml b/.github/workflows/build.yaml
index 6adb69563..13
Am 30.01.23 um 17:17 schrieb Gert Doering:
overlapped_io_init() has a "bool tuntap_buffer" argument which is only
passed onwards to alloc_buf_sock_tun(), which does nothing with it.
Remove from both functions.
While at it, move alloc_buf_sock_tun() from mtu.c to win32.c and make
static. It's o
overlapped_io_init() has a "bool tuntap_buffer" argument which is only
passed onwards to alloc_buf_sock_tun(), which does nothing with it.
Remove from both functions.
While at it, move alloc_buf_sock_tun() from mtu.c to win32.c and make
static. It's only ever called from win32.c / overlapped_io_
overlapped_io_init() has a "bool tuntap_buffer" argument which is only
passed onwards to alloc_buf_sock_tun(), which does nothing with it.
Remove from both functions.
Signed-off-by: Gert Doering
---
src/openvpn/mtu.c| 3 +--
src/openvpn/mtu.h| 3 +--
src/openvpn/socket.c | 4 ++--
src/o
The undefined behaviour USAN clang checker found these two cases. The optimiser
of clang/gcc will optimise the memcpy away in the auth_token case and output
excactly the same assembly on amd64/arm64 but it is still better to not rely
on undefined behaviour.
The hw addr fix is a mess but so are the
The -fno-sanitize-recover=all flag ensures that for all errors we actually
abort the tests in the automated testing and not just print some errors in red
that nobody sess. Also add the undefined tests to catch more bugs
Signed-off-by: Arne Schwabe
---
.github/workflows/build.yaml | 10 +-
On Fri, Jan 27, 2023 at 08:41:38PM +0100, Matthias Andree wrote:
> Am 25.01.23 um 20:50 schrieb Frank Lichtenheld:
> > The OpenVPN community project team is proud to release OpenVPN 2.6.0.
> > This is the new stable version of OpenVPN with some major new features.
>
> Hi Frank,
>
> OpenVPN 2.5.x
11 matches
Mail list logo
