Re: netgraph panics

[Hiten Pandya <hmp@xxxxxxxxxxxxx>]

YONETANI Tomokazu wrote:
> On Mon, Aug 22, 2005 at 01:34:50PM +0100, Hiten Pandya wrote:
> When it happens, I can get rid of this anti-foot-shooting which has been
> in my tree for more than a year :)

OK, nice work Tomokazu-san.  That's a nice stopgap, but read on for my 
suggestion of a commitable workaround.

This is a good workaround instead of crying panic and not working at all. 
 One more thing you could do is load the ng_ksocket.ko module from 
kldload(2) system call's context in the module handler, which will then 
remove the requirement for the user to load this module.

Since the module handler is GUARANTEED to have a process context, you can 
easily call linker_preload_file from there, and load ng_ksocket.ko.

I mean, you could load ng_ksocket.ko in modhandler, ngb_mod_event() and 
you can KEEP this ifdef section in ng_mkpeer so that OTHER netgraph module 
files do not fall into this trap.  So under the MOD_LOAD case, you would 
call the linker_preload_file() function, and unload it under the 
MOD_UNLOAD case, naturally.

By adding such a workaround, we at least know that the basic netgraph 
operations will not fail because ng_ksocket.ko is missing, since we 
preloaded one of the fundamental ng_ modules for them; they only need to 
worry about the specific ones.

If you can develop upon your foot-shooting measure with the above 
suggestion, you can go ahead and commit it! :-)

Thanks!

				Hiten Pandya
				hmp@xxxxxxxxxxxxx