Re: [PATCH bpf-next 3/5] bpf: Mark linux_binprm->mm as trusted-or-null

> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> index add3affc57035..8bdcb65618d01 100644
> --- a/kernel/bpf/verifier.c
> +++ b/kernel/bpf/verifier.c
> @@ -5682,6 +5682,10 @@ BTF_TYPE_SAFE_TRUSTED_OR_NULL(struct dentry) {
>  	struct inode *d_inode;
>  };
>
> +BTF_TYPE_SAFE_TRUSTED_OR_NULL(struct linux_binprm) {
> +	struct mm_struct *mm;
> +};
> +

Does this break the BPF load-compatibility guarantee documented in
Documentation/bpf/bpf_design_QA.rst?

Before this patch, walking bprm->mm produced a dereferenceable register
in every program type. In sleepable programs, in_rcu_cs() is false, so
check_ptr_to_btf_access() sets flag = PTR_UNTRUSTED. In non-sleepable
programs, in_rcu_cs() is true but mm is not in any __safe_rcu* list, so
clear_trusted_flags() produces a plain PTR_TO_BTF_ID.

Both PTR_TO_BTF_ID and PTR_TO_BTF_ID|PTR_UNTRUSTED satisfy the guard in
check_mem_access():

  } else if (base_type(reg->type) == PTR_TO_BTF_ID &&
             !type_may_be_null(reg->type)) {
          err = check_ptr_to_btf_access(env, regs, reg, argno, off, size, t,
                                        value_regno);

so bprm->mm->arg_start loads with no NULL check.

After this patch, btf_nested_type_is_trusted() matches
linux_binprm__safe_trusted_or_null.mm, so check_ptr_to_btf_access() sets
flag |= PTR_TRUSTED | PTR_MAYBE_NULL. Now type_may_be_null() is true, the
arm above is skipped, and control falls to the final else:

  } else {
          verbose(env, "%s invalid mem access '%s'\n", reg_arg_name(env, argno),
                  reg_type_str(env, reg->type));
          return -EACCES;
  }

Any existing program that reads through bprm->mm without an explicit NULL
check now fails bpf_prog_load() with -EACCES.

> @@ -5732,6 +5736,7 @@ static bool type_is_trusted_or_null(struct bpf_verifier_env *env,
>  {
>  	BTF_TYPE_EMIT(BTF_TYPE_SAFE_TRUSTED_OR_NULL(struct socket));
>  	BTF_TYPE_EMIT(BTF_TYPE_SAFE_TRUSTED_OR_NULL(struct dentry));
> +	BTF_TYPE_EMIT(BTF_TYPE_SAFE_TRUSTED_OR_NULL(struct linux_binprm));
>  	BTF_TYPE_EMIT(BTF_TYPE_SAFE_TRUSTED_OR_NULL(struct vm_area_struct));
>
>  	return btf_nested_type_is_trusted(&env->log, reg, field_name, btf_id,
> diff --git a/tools/testing/selftests/bpf/progs/lsm.c b/tools/testing/selftests/bpf/progs/lsm.c
> index 7de173daf27b6..7441d66c080c9 100644
> --- a/tools/testing/selftests/bpf/progs/lsm.c
> +++ b/tools/testing/selftests/bpf/progs/lsm.c
> @@ -113,6 +113,7 @@ int BPF_PROG(test_void_hook, struct linux_binprm *bprm)
>  {
>  	__u32 pid = bpf_get_current_pid_tgid() >> 32;
>  	struct inner_map *inner_map;
> +	struct mm_struct *mm;
>  	char args[64];
>  	__u32 key = 0;
>  	__u64 *value;
> @@ -121,7 +122,9 @@ int BPF_PROG(test_void_hook, struct linux_binprm *bprm)
>  		bprm_count++;
>
>  	bpf_copy_from_user(args, sizeof(args), (void *)bprm->vma->vm_mm->arg_start);
> -	bpf_copy_from_user(args, sizeof(args), (void *)bprm->mm->arg_start);
> +	mm = bprm->mm;
> +	if (mm)
> +		bpf_copy_from_user(args, sizeof(args), (void *)mm->arg_start);
>
>  	value = bpf_map_lookup_elem(&array, &key);
>  	if (value)

The selftest rewrite above shows the problem: the old form
bpf_copy_from_user(args, sizeof(args), (void *)bprm->mm->arg_start) now
requires a NULL check. Out-of-tree BPF LSM and tracing programs that read
bprm->mm the same way will stop loading on kernel upgrade with no source
change on their side.

The bpf_design_QA.rst compatibility guarantee states: "The bpf development
process guarantees that the future kernel versions will accept all bpf
programs that were accepted by the earlier versions."

The tracing-program exception in that document does not cover this case:
struct linux_binprm's layout is unchanged, the program source is unchanged,
and the in-tree casualty is a BPF LSM program, not a bpf_probe_read()-based
tracing program.

Should the changelog mention the compatibility impact and document that
programs dereferencing bprm->mm now need a NULL check?


---
AI reviewed your patch. Please fix the bug or email reply why it's not a bug.
See: https://github.com/kernel-patches/vmtest/blob/master/ci/claude/README.md

CI run summary: https://github.com/kernel-patches/bpf/actions/runs/31488001112