Re: [ros-svn] [ion] 14884: Add hack for ROS's weird behavior. Will investigate but this lets you boot for now

[Alex Ionescu]

[e-mail] wrote:
> Add hack for ROS's weird behavior. Will investigate but this lets you boot for now
>
> Updated files:
> trunk/reactos/ntoskrnl/io/irp.c

The problem is that a bunch of places in the code set the FileObject 
Event as the UserEvent. This is incredibly wrong, because both are used 
to signal completely different things and shouldn't be used like that. 
I've changed them to local stack KEVENTS, but this still fails, because 
you cannot dereference them. The check for !IRP_SYNCHRONOUS_API is meant 
to ensure against this (Syncronous IRPs have KEVENTS, Async ones have 
Executive Events, in short terms.) I've researched some more and this is 
because of a bigger problem. The routines doing the hack all assumed 
that the File Object is Synchronous and build a Sync FSD (with 
IRP_SYNCH_API set), and then set a KEVENT UserEvent. They should 
actually check if the File Object is Sync or Async, and in case of Async 
they should use the Local Stack KEVENT (which won't be dereferenced 
because of the check in IoCompleteRequest APC (2nd stage)), while in 
case of async they shouldn't use any event at all and wait on the file 
object instead of waiting on the local event. Also, in case of a sync 
operation, we must eventually include some sort of locking/unlocking of 
the file object for serialized access which is required. I will spend 
some time tomorrow working with these issues so the hack can be removed 
and proper kernel functionality to be restored, but I wanted to share 
the background information.

Best regards,
Alex Ionescu