Re: [PATCH 2/6] ntoskrnl.exe: Implement KeReleaseSemaphore() and waiting on semaphores.

[Zebediah Figura]

On 11/27/18 12:58 AM, Dmitry Timoshkov wrote:
> Zebediah Figura  wrote:
> > -
> >        case TYPE_SEMAPHORE:
> > -
> >        {
> > -
> >            KSEMAPHORE *semaphore = CONTAINING_RECORD(objs[i], KSEMAPHORE, Header);
> > -
> >            objs[i]->WaitListHead.Blink = CreateSemaphoreW( NULL,
> > -
> >                semaphore->Header.SignalState, semaphore->Limit, NULL );
> > -
> >            break;
> > -
> >        }
>
> ...
> > - EnterCriticalSection( &sync_cs );
> > - ret = InterlockedExchangeAdd( &semaphore->Header.SignalState, count );
> > - if (handle)
> > -
> >    ReleaseSemaphore( handle, count, NULL );
> > - LeaveCriticalSection( &sync_cs );
>
> Wouldn't it be better to use native instead of win32 APIs? This way you would
> be able to propagate errors instead of ignoring them.

Kernel sync functions, as a rule, don't return errors. For example, 
KeReleaseSemaphore() returns the previous count, and raises an exception 
if the limit is exceeded. Similarly, while KeWaitForMultipleObjects() 
does return an NTSTATUS, error conditions (like passing more than 64 
objects, or waiting too many times on a mutex) result in exceptions or 
panics.