[PATCH bpf-next RESEND v2 1/2] bpf: Take return from set_memory_ro() into account with bpf_prog_lock_ro()

[Daniel Borkmann]

On 3/1/24 8:57 AM, Christophe Leroy wrote:
> set_memory_ro() can fail, leaving memory unprotected.
> 
> Check its return and take it into account as an error.
> 
> Link: https://github.com/KSPP/linux/issues/7
> Signed-off-by: Christophe Leroy <christophe.leroy at csgroup.eu>
> Cc: linux-hardening at vger.kernel.org <linux-hardening at vger.kernel.org>
> Reviewed-by: Kees Cook <keescook at chromium.org>
> ---
> Sorry for the resend, I forgot to flag patch 2 as bpf-next
> 
> Note: next patch is autonomous, it is sent as a follow-up of this one to minimize risk of conflict on filter.h because the two changes are too close to each other.
> 
> v2: No modification (Just added link in patch message), patchwork discarded this series due to failed test of s390 but it seems unrelated, see https://lore.kernel.org/bpf/wvd5gzde5ejc2rzsbrtwqyof56uw5ea3rxntfrxtkdabzcuwt6@w7iczzhmay2i/T/#m2e61446f42d5dc3d78f2e0e8b7a783f15cfb109d
> ---
>   include/linux/filter.h | 5 +++--
>   kernel/bpf/core.c      | 4 +++-
>   kernel/bpf/verifier.c  | 4 +++-
>   3 files changed, 9 insertions(+), 4 deletions(-)
> 
> diff --git a/include/linux/filter.h b/include/linux/filter.h
> index 36cc29a2934c..7dd59bccaeec 100644
> --- a/include/linux/filter.h
> +++ b/include/linux/filter.h
> @@ -884,14 +884,15 @@ bpf_ctx_narrow_access_offset(u32 off, u32 size, u32 size_default)
>   
>   #define bpf_classic_proglen(fprog) (fprog->len * sizeof(fprog->filter[0]))
>   
> -static inline void bpf_prog_lock_ro(struct bpf_prog *fp)
> +static inline int __must_check bpf_prog_lock_ro(struct bpf_prog *fp)
>   {
>   #ifndef CONFIG_BPF_JIT_ALWAYS_ON
>   	if (!fp->jited) {
>   		set_vm_flush_reset_perms(fp);
> -		set_memory_ro((unsigned long)fp, fp->pages);
> +		return set_memory_ro((unsigned long)fp, fp->pages);
>   	}
>   #endif
> +	return 0;
>   }
>   
>   static inline void bpf_jit_binary_lock_ro(struct bpf_binary_header *hdr)
> diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c
> index 71c459a51d9e..c49619ef55d0 100644
> --- a/kernel/bpf/core.c
> +++ b/kernel/bpf/core.c
> @@ -2392,7 +2392,9 @@ struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err)
>   	}
>   
>   finalize:
> -	bpf_prog_lock_ro(fp);
> +	*err = bpf_prog_lock_ro(fp);
> +	if (*err)
> +		return fp;
>   
>   	/* The tail call compatibility check can only be done at
>   	 * this late stage as we need to determine, if we deal
> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> index 1c34b91b9583..6ec134f76a11 100644
> --- a/kernel/bpf/verifier.c
> +++ b/kernel/bpf/verifier.c
> @@ -19096,7 +19096,9 @@ static int jit_subprogs(struct bpf_verifier_env *env)
>   	 * bpf_prog_load will add the kallsyms for the main program.
>   	 */
>   	for (i = 1; i < env->subprog_cnt; i++) {
> -		bpf_prog_lock_ro(func[i]);
> +		err = bpf_prog_lock_ro(func[i]);
> +		if (err)
> +			goto out_free;

How does the error path take out the subprogs from kallsyms in your case? Suppose some of
the loop iterations succeed before we hit an error. I believe the subprogs still exist in
kallsyms here.

>   		bpf_prog_kallsyms_add(func[i]);
>   	}
>   
>