Re: [PATCH bpf-next v5 4/7] bpf: lsm: Implement attach, detach and execution

[Yonghong Song <yhs@fb.com>]

On 3/23/20 9:44 AM, KP Singh wrote:
> From: KP Singh <kpsingh@google.com>
> 
> JITed BPF programs are dynamically attached to the LSM hooks
> using BPF trampolines. The trampoline prologue generates code to handle
> conversion of the signature of the hook to the appropriate BPF context.
> 
> The allocated trampoline programs are attached to the nop functions
> initialized as LSM hooks.
> 
> BPF_PROG_TYPE_LSM programs must have a GPL compatible license and
> and need CAP_SYS_ADMIN (required for loading eBPF programs).
> 
> Upon attachment:
> 
> * A BPF fexit trampoline is used for LSM hooks with a void return type.
> * A BPF fmod_ret trampoline is used for LSM hooks which return an
>    int. The attached programs can override the return value of the
>    bpf LSM hook to indicate a MAC Policy decision.
> 
> Signed-off-by: KP Singh <kpsingh@google.com>
> Reviewed-by: Brendan Jackman <jackmanb@google.com>
> Reviewed-by: Florent Revest <revest@google.com>
> ---
>   include/linux/bpf.h     |  4 ++++
>   include/linux/bpf_lsm.h | 11 +++++++++++
>   kernel/bpf/bpf_lsm.c    | 29 +++++++++++++++++++++++++++++
>   kernel/bpf/btf.c        |  9 ++++++++-
>   kernel/bpf/syscall.c    | 26 ++++++++++++++++++++++----
>   kernel/bpf/trampoline.c | 17 +++++++++++++----
>   kernel/bpf/verifier.c   | 19 +++++++++++++++----
>   7 files changed, 102 insertions(+), 13 deletions(-)
> 
> diff --git a/include/linux/bpf.h b/include/linux/bpf.h
> index af81ec7b783c..adf2e5a6de4b 100644
> --- a/include/linux/bpf.h
> +++ b/include/linux/bpf.h
> @@ -433,6 +433,10 @@ struct btf_func_model {
>    * programs only. Should not be used with normal calls and indirect calls.
>    */
>   #define BPF_TRAMP_F_SKIP_FRAME		BIT(2)
> +/* Override the return value of the original function. This flag only makes
> + * sense for fexit trampolines.
> + */
> +#define BPF_TRAMP_F_OVERRIDE_RETURN     BIT(3)

Whether the return value is overridable is determined by hook return 
type as below. Do we still need this flag?

>   
>   /* Each call __bpf_prog_enter + call bpf_func + call __bpf_prog_exit is ~50
>    * bytes on x86.  Pick a number to fit into BPF_IMAGE_SIZE / 2
[...]