On Fri, Dec 12, 2008 at 07:26:10AM -0700, Robert Thurlow wrote:
> Nicolas Williams wrote:
> >where automountd_wait_for_cleanup() just waits for the child and then
> >exits. The parent there seems to exist to serve door requests on
> >did_fork_exec and did_exec_map, but why must that be done in a separate
> >process when both processes run with the same credentials and
> >privileges?
>
> This part is due to hangs we had when using executable
> maps (ick). See 6368753 for more.
Fascinating. Truly fascinating.
IMO auto_wait4mount() shouldn't sigintr() (one of the various linked CRs
had a similar comment), and maybe automountd should do the lookup in the
same thread and then do the mount(2) instead of this convoluted trigger-
a-lookup-which-upcalls-to-the-same-process business. But the fix of
having a separate process to do all executable map work was definitely
expedient. There's only the tiniest comment in autod_main.c that only
alludes to this ("This works around a potential hang in using fork1()
within a multithreaded environment").
Nico
--
