Here's the header file for the the proposed new ioctl interface for
dm. We've tried to change as little as possible to minimise code
changes in LVM2 and EVMS.
The changes address the following problems:
o) Alignment mistakes in the ioctl structures
o) Provide a way for the userland tools to back out of a series of
table updates without having to reload the orginal tables (remember
that an LV is often composed of a stack of 2 or more dm devices).
o) Pre-loading of a new table may now occur without suspending the
device. (Table loading can potentially allocate a lot of memory,
eg. mirror buffers).
o) Minimise the amount of time that a device is suspended. eg. if a
simple linear mapping is being extended, there is no longer a need to
call the suspend ioctl in between the table load and the resume.
Instead the resume (which swaps in the new table), will do the right
thing.
- Joe
This archive was generated by hypermail 2b29 : Sat Jun 07 2003 - 22:00:26 EST