Looks like its getting there!
Right, how do we sort out a _severe_ deficiency in the BSD implementation?
The deficiency I refer is
1) get fd of some mmap()able device
2) mmap it into our address space
3) oh, its been revoke()'d from us
4) never mind! We can read/write it because we have it mapped :) :)
Granted, mmap() doesn't work for many block/char devs., but quick /dev/mem
tests on hastily-installed FreeBSD showed the problem.
In the above scenario, can the suspicously useful sounding inode->i_mmap
help us in any way, to forcibly unmap any areas based on a revoke()'d fd?
Cheers
Chris