Re: Thoughts about proc permission handling

[Dave Hudson <dave@humbug.demon.co.uk> writes:]

>Robert Mayer - Student wrote:
>> If race conditions are acceptable, why not keep pstat() a system call
>> but make it retrieve the information without doing any locking ?
>It rather depends who we race with - the danger is that our underlying
>process structures can disappear from under us if we don't make sure we're
>the only one playing with them.

Think even further out. If a user program gets befuddled, he might take a
segv and die. The system quite likely goes on. If the kernel gets
befuddled....

Also, put on your microkernel hat. If it doesn't need to go in the kernel,
don't put it there. Keep your ring 0 code small, allow as much code as
possible to run in an environment where faults can be contained. Also,
saddle the exceptional case with the cost. The cost of making queries about
process state should be skewed heavily to the process making the query.

                                                        Andy
Received on Mon Dec 5 10:37:31 1994