[PATCH] root-hopping for pre-2.3.41-3

[almesber en lrc.di.epfl.ch]

Horst von Brand wrote:
> I'd say go with the proposal of the fake root/cwd directory for kernel
> threads that don't need access to the filesystem (all of them, I
> guess). Cleanest way out, unless you allow them to live without root/cwd.

I agree with clean. The main problem here is to catch all threads/processes, 
including the ones not (yet) in the mainstream kernel and those which people 
haven't even written yet. I don't want something that breaks every once in a
while in subtle ways for some silly and apparently unrelated reason.

Of course there's The One Right Way(tm) to do it:

 - rename _*kernel_thread to _*kernel_thread_with_fs
 - find all other possible paths leading to kernel_thread/clone and do 
   the same thing there
 - construct the dentry jail (may be a little tricky, e.g. I'm not sure
   if d_inode = NULL is safe, and there are a few other pointers I don't
   quite trust)
 - write a kernel_thread_without_fs that calls kernel_thread_with_fs
   and then puts the process into the dentry jail
 - change all 30+ calls to _*kernel_thread to 
   _*kernel_thread_without_fs (and repeat this for all the other paths 
   you've found above)
 - notify the authors, so they know what's happening to their code

This still has the following problems:

 - you've just broken a few externally maintained packages, e.g.
   probably PCMCIA
 - threads that start early but access the file system only later
   still cause problems
 - doesn't help a bit with processes that aren't kernel threads

Of course, the latter two are somewhat hypothetical at the moment.

- Werner

-- 
  _________________________________________________________________________
 / Werner Almesberger, ICA, EPFL, CH       werner.almesberger en ica.epfl.ch /
/_IN_N_032__Tel_+41_21_693_6621__Fax_+41_21_693_6610_____________________/

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo en vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/