Re: [PATCH v2] binder: use buffer offsets in debug logs

[Tiffany Y. Yang]

Carlos Llamas <cmllamas@xxxxxxxxxx> writes:

> On Tue, Mar 25, 2025 at 12:41:39AM +0000, Tiffany Y. Yang wrote:
>> Carlos Llamas <cmllamas@xxxxxxxxxx> writes:
>> 
>> > On Mon, Mar 24, 2025 at 06:07:18PM +0000, Tiffany Y. Yang wrote:
>> >> Identify buffer addresses using vma offsets instead of full user
>> >> addresses in debug logs.
>> >> 
>> >> Signed-off-by: Tiffany Y. Yang <ynaffit@xxxxxxxxxx>
>> >> ---
>> >>  drivers/android/binder.c | 31 ++++++++++++++++---------------
>> >>  1 file changed, 16 insertions(+), 15 deletions(-)
>> >> 
>> >> diff --git a/drivers/android/binder.c b/drivers/android/binder.c
>> >> index d1aa6d24450a..994ae205aa07 100644
>> >> --- a/drivers/android/binder.c
>> >> +++ b/drivers/android/binder.c
>> >> @@ -3261,20 +3261,20 @@ static void binder_transaction(struct binder_proc *proc,
>> >>  
>> >>  	if (reply)
>> >>  		binder_debug(BINDER_DEBUG_TRANSACTION,
>> >> -			     "%d:%d BC_REPLY %d -> %d:%d, data %016llx-%016llx size %lld-%lld-%lld\n",
>> >> +			     "%d:%d BC_REPLY %d -> %d:%d, buffer offset %lx-%lx size %lld-%lld-%lld\n",
>> >>  			     proc->pid, thread->pid, t->debug_id,
>> >>  			     target_proc->pid, target_thread->pid,
>> >> -			     (u64)tr->data.ptr.buffer,
>> >> -			     (u64)tr->data.ptr.offsets,
>> >> +			     (unsigned long)tr->data.ptr.buffer - proc->alloc.buffer,
>> >> +			     (unsigned long)tr->data.ptr.offsets - proc->alloc.buffer,
>> >
>> > These could be pointers to anywhere in user memory, not necessarily the
>> > alloc->buffer. So there will be cases where this substraction doesn't
>> > make sense. However, you are correct that we shouldn't log these addrs
>> > so maybe just don't? wdyt?
>> >
>> 
>> Ah, in that case I think it makes sense to remove them here. What
>> do you think about printing the full buffer and offsets values in cases
>> where we would print a binder_user_error or binder_transaction_error
>> instead. Ideally, I would try to limit this to cases when the data or
>> offsets ptr is invalid / copy would fail. Ostensibly this wouldn't
>> reveal dangerous information about the user address space because the
>> print statements would only happen when the data wasn't where it was
>> supposed to be and it would help with debugging, but I'm not sure if
>> this line of thought makes sense...
>
> My 2 cents...
>
> I'm sure there will be a _few_ exceptions in which having the pointers
> from binder_transaction_data logged would aid debugging. However, this
> won't be info that most users care about. In practice, logging an error
> with "invalid buffer/offsets pointer" message is enough.
>
> There are _other_ pointers that users do care about when debugging, such
> as binder_ptr_cookie but not these. So I think is better if we don't log
> them at all, as calculating an "offset" is not possible either.

This makes sense to me! I'll drop them and send out another patch.