On 4/3/26 00:32, Alexei Starovoitov wrote:

On Wed, Feb 11, 2026 at 7:13 AM Leon Hwang [off-list ref] wrote:

quoted

[...]

quoted

@@ -6241,7 +6244,11 @@ static int __sys_bpf(enum bpf_cmd cmd, bpfptr_t uattr, unsigned int size,
                err = map_freeze(&attr);
                break;
        case BPF_PROG_LOAD:
-               err = bpf_prog_load(&attr, uattr, size);
+               if (from_user && size >= offsetofend(union bpf_attr, log_true_size))
+                       log_true_size = uattr.user + offsetof(union bpf_attr, log_true_size);

So you added 'from_user' gating because
you replaced copy_to_bpfptr_offset() with copy_to_user()?
This is a drastic change in behavior and you don't even talk about
it in the commit log.
You said "refactor". This is not a refactoring!

This is v10. The common_attr feature is useful, but
you really need to think harder about what your patches
are doing.

Refactoring should not introduce any functional changes. If a functional
change is involved, it should be factored out of the refactoring commit
into a separate commit with an explanation in the commit log.

I'll add this to my self-review checklist.

The intention of 'from_user' was to replace copy_to_bpfptr_offset() with
copy_to_user(), since the log is always copied to the user-space buffer
when the log level is not BPF_LOG_KERNEL in
kernel/bpf/log.c::bpf_verifier_vlog().

The 'from_user' gating will be dropped in v12 to keep this patch as pure
refactoring.

Thanks,
Leon