Re: Write Order Restrictions

Can Sar [off-list ref] wrote:

the driver just cycles through all devices that make up a soft raid 
device and just calls generic_make_request on them. Is this correct, or 
does some other function involved in the write process (starting from 
the soft raid level down) actually wait on buffers to enforce 
restrictions that e.g. cause blocks for one disk to be written before 
those of another?

None that I know of.  Neil may have more to say.

In particular, I don't know of any *explicit* mechanism that guarantees
that write order is preserved on any ONE mirror (in the case of raid1).

On the other hand, md interposes its own make_request (as I recall),
so the generic kernel request elevator aggregation algorithm does not
apply.  And I think I remember that it does not allow queues greater
than length 1. If so, that would be helpful.

Peter