Re: [PATCH bpf-next 2/2] bpf: Take return from set_memory_rox() into account with bpf_jit_binary_lock_ro()

[Kees Cook <keescook@chromium.org>]

On Sun, Feb 18, 2024 at 11:55:02AM +0100, Christophe Leroy wrote:
> set_memory_rox() can fail, leaving memory unprotected.
> 
> Check return and bail out when bpf_jit_binary_lock_ro() returns
> and error.
> 
> Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
> ---
> Previous patch introduces a dependency on this patch because it modifies bpf_prog_lock_ro(), but they are independant.
> It is possible to apply this patch as standalone by handling trivial conflict with unmodified bpf_prog_lock_ro().
> ---
>  arch/arm/net/bpf_jit_32.c        | 25 ++++++++++++-------------
>  arch/arm64/net/bpf_jit_comp.c    | 21 +++++++++++++++------
>  arch/loongarch/net/bpf_jit.c     | 21 +++++++++++++++------
>  arch/mips/net/bpf_jit_comp.c     |  3 ++-
>  arch/parisc/net/bpf_jit_core.c   |  8 +++++++-
>  arch/s390/net/bpf_jit_comp.c     |  6 +++++-
>  arch/sparc/net/bpf_jit_comp_64.c |  6 +++++-
>  arch/x86/net/bpf_jit_comp32.c    |  3 +--
>  include/linux/filter.h           |  4 ++--
>  9 files changed, 64 insertions(+), 33 deletions(-)
> 
> diff --git a/arch/arm/net/bpf_jit_32.c b/arch/arm/net/bpf_jit_32.c
> index 1d672457d02f..01516f83a95a 100644
> --- a/arch/arm/net/bpf_jit_32.c
> +++ b/arch/arm/net/bpf_jit_32.c
> @@ -2222,28 +2222,21 @@ struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog)
>  	/* If building the body of the JITed code fails somehow,
>  	 * we fall back to the interpretation.
>  	 */
> -	if (build_body(&ctx) < 0) {
> -		image_ptr = NULL;
> -		bpf_jit_binary_free(header);
> -		prog = orig_prog;
> -		goto out_imms;
> -	}
> +	if (build_body(&ctx) < 0)
> +		goto out_free;
>  	build_epilogue(&ctx);
>  
>  	/* 3.) Extra pass to validate JITed Code */
> -	if (validate_code(&ctx)) {
> -		image_ptr = NULL;
> -		bpf_jit_binary_free(header);
> -		prog = orig_prog;
> -		goto out_imms;
> -	}
> +	if (validate_code(&ctx))
> +		goto out_free;
>  	flush_icache_range((u32)header, (u32)(ctx.target + ctx.idx));
>  
>  	if (bpf_jit_enable > 1)
>  		/* there are 2 passes here */
>  		bpf_jit_dump(prog->len, image_size, 2, ctx.target);
>  
> -	bpf_jit_binary_lock_ro(header);
> +	if (bpf_jit_binary_lock_ro(header))
> +		goto out_free;
>  	prog->bpf_func = (void *)ctx.target;
>  	prog->jited = 1;
>  	prog->jited_len = image_size;
> @@ -2260,5 +2253,11 @@ struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog)
>  		bpf_jit_prog_release_other(prog, prog == orig_prog ?
>  					   tmp : orig_prog);
>  	return prog;
> +
> +out_free:
> +	image_ptr = NULL;
> +	bpf_jit_binary_free(header);
> +	prog = orig_prog;
> +	goto out_imms;

These gotos give me the creeps, but yes, it does appear to be in the
style of the existing error handling.

> [...]
> diff --git a/arch/x86/net/bpf_jit_comp32.c b/arch/x86/net/bpf_jit_comp32.c
> index b18ce19981ec..f2be1dcf3b24 100644
> --- a/arch/x86/net/bpf_jit_comp32.c
> +++ b/arch/x86/net/bpf_jit_comp32.c
> @@ -2600,8 +2600,7 @@ struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog)
>  	if (bpf_jit_enable > 1)
>  		bpf_jit_dump(prog->len, proglen, pass + 1, image);
>  
> -	if (image) {
> -		bpf_jit_binary_lock_ro(header);
> +	if (image && !bpf_jit_binary_lock_ro(header)) {

I find the "!" kind of hard to read the "inverted" logic (0 is success),
so if this gets a revision, maybe do "== 0"?:

	if (image && bpf_jit_binary_lock_ro(header) == 0) {

But that's just me. So, regardless:

Reviewed-by: Kees Cook <keescook@chromium.org>

-- 
Kees Cook