[RFC] change of ->bd_op->open() semantics

[Alexander Viro]

	Folks, I propose to cache the result of blk_get_queue() in
struct block_device upon ->open().

	We already have struct block_device * whenever we are looking for
a queue.  So the only real problem arises if value of blk_get_queue()
changes while device is opened.  AFAICS nothing in the kernel pulls
such tricks (and to support them we would need to take care of a lot
of things, so changing the cached value wouldn't be a problem anyway).

	Yes, it means updating every instance of ->open() for block
devices.  All 30-odd of them.  I can do that in a single pass - changes
are trivial.

	It has an additional benefit of killing the array of default
queues on the same pass - a thing we will need to do sooner or later
anyway.

	Does anybody see a problem with that?