Signed-off-by: Evan Lezar <elezar@nvidia.com>

[no-relnote] Add SECURITY.md to repo

Signed-off-by: Jamie Liu <jamieliu@google.com>

Do not discard container flags when --cuda-compat-mode is not specified

In the case where no cuda-compat-mode was specified on the nvidia-container-cli configure commandline, the calculated set of flags were being CLEARED instead of setting a single bit.

This change ensures that the correct bit is set and the remaining bits are left unchanged.

This was not affecting the nvidia-container-runtime-hook's usage of the nvidia-container-cli since we explicitly add flags in that case.

This change adds additional logging around the flags and the
cuda-compat-mode option.

Signed-off-by: Evan Lezar <elezar@nvidia.com>

Add additional logging

The C11 standard states that the value of a pointer is undefined
outside of its lifetime. Since we were initilizing the argv pointer
in two conditional blocks, compilers that implement this standard were
cleaning up the value and this caused invalid dereferences of the
pointer.

The change introduces two local variables with the same lifetime of the
argv pointer and assigns these in the conditional blocks instead.

Signed-off-by: Evan Lezar <elezar@nvidia.com>

Fix pointer accessing local variable out of scope