On Wed, May 13, 2020 at 09:34:49AM +0200, Petr Mladek wrote:

On Tue 2020-05-12 11:45:54, Kees Cook wrote:

quoted

Here are the problems I see being solved by this:

- lifting kmsg dump reason filtering out of the individual pstore
  backends and making it part of the "infrastructure", so that
  there is a central place to set expectations. Right now there
  is a mix of explicit and implicit kmsg dump handling:

  - arch/powerpc/kernel/nvram_64.c has a hard-coded list

It handles restart, halt, poweroff the same way.  I wonder if anyone
would want to distinguish them.

quoted

  - drivers/firmware/efi/efi-pstore.c doesn't expect anything but
    OOPS and PANIC.
  - drivers/mtd/mtdoops.c tries to filter using its own dump_oops
    and doesn't expect anything but OOPS and PANIC.
  - fs/pstore/ram.c: has a hard-coded list and uses its own
    dump_oops.
  - drivers/mtd/mtdpstore.c (under development[3]) expected only
    OOPS and PANIC and had its own dump_oops.

The others handle only panic or oops.

What about splitting the reason into two variables? One for severity
and other for shutdown behavior. I mean:

  + reason: panic, oops, emergency, shutdown    (ordered by severity)
  + handling: restart, halt, poweroff

Or we might just replace KMSG_DUMP_RESTART, KMSG_DUMP_HALT,
KMSG_DUMP_POWEROFF with a single KMSG_DUMP_SHUTDOWN.

Then the max reason variable would make sense.

That would work for me, yeah. Pavel, is that enough granularity for you?

-- 
Kees Cook