Hi all,
what is the consensus on being able to use e.g. iostat(8) on
a kernel dump? This is currently possible, because the implementation
is kvm(4) based and it is therefore free. I'd like to move this and
other tools in the tree as well (like ps) to the full sysctl-driven
libkinfo. Beside removing the need for setgid kmem, this also
consolidates the access to kernel structures. Ideally, the need for
recompilation this userland tools will completly vanish.
This does come with a cost and that's the ability to work on core dump.
I don't think this is a big deal, because this can be done via gdb
instead, but I don't like to remove functionality without at least
a warning.
Another interesting question is the importance of the lastpid field
top(1) shows. This is based on the nextpid kernel variable and it
can argued that exposing this e.g. for jails is a minor information
leak. Do we want to keep this functionality? Should jails show this?
Joerg