Thomas Gleixner <tglx@xxxxxxxxxxxxx> writes:
> On Mon, Oct 21 2024 at 09:46, Björn Töpel wrote:
>> Celeste Liu <coelacanthushex@xxxxxxxxx> writes:
>>> 1. syscall_enter_from_user_mode() will do two things:
>>> 1) the return value is only to inform whether the syscall should be skipped.
>>> 2) regs will be modified by filters (seccomp or ptrace and so on).
>>> 2. for common entry user, there is two informations: syscall number and
>>> the return value of syscall_enter_from_user_mode() (called is_skipped below).
>>> so there is three situations:
>>> 1) if syscall number is invalid, the syscall should not be performed, and
>>> we set a0 to -ENOSYS to inform userspace the syscall doesn't exist.
>>> 2) if syscall number is valid, is_skipped will be used:
>>> a) if is_skipped is -1, which means there are some filters reject this syscall,
>>> so the syscall should not performed. (Of course, we can use bool instead to
>>> get better semantic)
>>> b) if is_skipped != -1, which means the filters approved this syscall,
>>> so we invoke syscall handler with modified regs.
>>>
>>> In your design, the logical condition is not obvious. Why syscall_enter_from_user_mode()
>>> informed the syscall will be skipped but the syscall handler will be called
>>> when syscall number is invalid? The users need to think two things to get result:
>>> a) -1 means skip
>>> b) -1 < 0 in signed integer, so the skip condition is always a invalid syscall number.
>>>
>>> In may way, the users only need to think one thing: The syscall_enter_from_user_mode()
>>> said -1 means the syscall should not be performed, so use it as a condition of reject
>>> directly. They just need to combine the informations that they get from API as the
>>> condition of control flow.
>>
>> I'm all-in for simpler API usage! Maybe massage the
>> syscall_enter_from_user_mode() (or a new one), so that additional
>> syscall_get_nr() call is not needed?
>
> It's completely unclear to me what the actual problem is. The flow how
> this works on all architectures is:
>
> regs->orig_a0 = regs->a0
> regs->a0 = -ENOSYS;
>
> nr = syscall_enter_from_user_mode(....);
>
> if (nr >= 0)
> regs->a0 = nr < MAX_SYSCALL ? syscall(nr) : -ENOSYS;
>
> If syscall_trace_enter() returns -1 to skip the syscall, then regs->a0
> is unmodified, unless one of the magic operations modified it.
>
> If syscall_trace_enter() was not active (no tracer, no seccomp ...) then
> regs->a0 already contains -ENOSYS.
>
> So what's the exact problem?
It's a mix of calling convention, and UAPI:
* RISC-V uses a0 for arg0 *and* return value (like arm64).
* RISC-V does not expose orig_a0 to userland, and cannot easily start
doing that w/o breaking UAPI.
Now, when setting a0 to -ENOSYS, it's clobbering arg0, and the ptracer
will have an incorrect arg0 (-ENOSYS).
