Re: zvol device weirdness after memory corruption?

[mlelstv%serpens.de@localhost (Michael van Elst)]

gdt%lexort.com@localhost (Greg Troxel) writes:

>I am somewhat surprised that ccd uses the block device but that's what
>the man page says.

Almost everything in the kernel uses the block device as the char
device (or rather the physio() function used by it) supports I/O
only from userland.

An exception is the wedge discovery mechanism. Since there can be
only a single opener of a block device, it needs to work around
this by:
- opening the char device.
- determining the corresponding block device.
- calling the block device strategy routine for I/O.
- closing the char device again.

The device handling and arbitrating mechanisms are pretty arcane.