Re: [PATCH bpf-next v4 2/3] bpf: implement dynptr copy kfuncs

[Mykyta Yatsenko <mykyta.yatsenko5@gmail.com>]

On 5/9/25 16:50, Alexei Starovoitov wrote:
> On Thu, May 8, 2025 at 3:06 PM Mykyta Yatsenko
> <mykyta.yatsenko5@gmail.com> wrote:
>> +static __always_inline int __bpf_dynptr_copy_str(struct bpf_dynptr *dptr, u32 doff, u32 size,
>> +                                                const void __user *unsafe_src,
>> +                                                copy_fn_t str_copy_fn,
>> +                                                struct task_struct *tsk)
>> +{
> ...
>> +__bpf_kfunc int bpf_copy_from_user_task_str_dynptr(struct bpf_dynptr *dptr, u32 off,
>> +                                                  u32 size, const void *unsafe_ptr__ign,
>> +                                                  struct task_struct *tsk)
>> +{
>> +       return __bpf_dynptr_copy_str(dptr, off, size, unsafe_ptr__ign,
>> +                                    copy_user_str_sleepable, tsk);
>> +}
> CI is not happy about implicit cast that changes address spaces:
>
> ../kernel/trace/bpf_trace.c:3702:55: warning: incorrect type in
> argument 4 (different address spaces)
> ../kernel/trace/bpf_trace.c:3702:55:    expected void const [noderef]
> __user *unsafe_src
> ../kernel/trace/bpf_trace.c:3702:55:    got void const *unsafe_ptr__ign
>
> Please use gcc 14 or higher or sparse to see them.
>
> Probably __bpf_dynptr_copy_str() shouldn't have __user qualifier,
> but bpf_copy_from_user_task_str_dynptr() should have __user next to
> unsafe_ptr__ign,
> and everywhere we kfunc has __user it can case to (const void *) before
> calling __bpf_dynptr_copy_str().
I'll check it, thanks.