On Sep 26, 2006, at 10:44 AM, Walter B. Ligon III wrote:


Hey guys, Phil, riddle me this ...

In pvfs2-server.c there is a function "server_state_machine_start_noreq()" which is used to start state machines that don't originate with a request message. When it starts the machine it passes a pointer to a local variable "tmp_status" in as job_status_p - which later gets passed to job functions in the state machine code ... which means that these pointers are potentially pointing to random places on the stack, I think.


We don't save those js_p variables anywhere in the job code. The state action functions take js_p variables, but they are passed in from the server code. The start_noreq calls are only used right now for perf updates and job timers. Both of those server op structures get stored in the job context when we do the start_noreq, but the job status pointers only get filled in by the call to job_testcontext.

-sam

I think this is a problem, and I intend to fix it ... just want to be sure I'm not missing something. I'm asking because I've seriously changed that stuff anyway and I'm not sure if the reason was obvious before, but not now, or what.

So, this leads to the next question ... about job_status_p's passed to job functions, I'm not entirely clear where is the best place to allocate and free them - they end up getting stuck in arrays and stuff int he job interface, and I'm finding it hard to follow their path. I assume the job interface will not tolerate a NULL pointer here (right?). I think this is just an issue because these "noreq" SMs are for things like timers that noone ever actually looks at the return status, but I want to be sure.

Comments appreciated.
Walt
--
Dr. Walter B. Ligon III
Associate Professor
ECE Department
Clemson University
_______________________________________________
Pvfs2-developers mailing list
[email protected]
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers


_______________________________________________
Pvfs2-developers mailing list
[email protected]
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers

Reply via email to