Re: [PATCH] exec: do not leave bprm->interp on stack

[P J P <ppandit@redhat.com>]

+-- On Mon, 19 Nov 2012, Kees Cook wrote --+
| I don't think you're being rude at all. You're defending your solution. :)

  Thank you Kees, really appreciate it.
 
| However, it also changes the conditions for when a module is loaded
| (i.e. 0x7f no longer triggers a module_load, so anything needing that
| would break -- I'm not sure if this really qualifies for ABI breakage,
| I don't use any obscure binfmt modules so I can't say).

   Ah right.
 
| And, most importantly, it triggers request_module for any binary with
| unprintables that binfmt_misc may already handle (for example, the
| very common case of handling DOS MZ files, which only define 2 bytes
| as magic (MZ) and exampes I find show things like "@\x00" trailing it,
| or JAR files which are PK\x03\x04). Which means each exec of these
| kinds of files would trigger a needless request_module() call on every
| exec.

   Hmmn...true.
 
| Both the interp-on-heap patch and this proposed ELOOP patch are needed
| to handle the case of binfmt_script and/or binfmt_misc being modules
| (first binfmt walk fails with -ENOEXEC, loads binfmt_script, retries
| loop, hits binfmt_script rewriting interp to a PE file, recurses,
| fails with -ENOEXEC, loads binfmt_misc via a modalias for PE files,
| retries loop, hits binfmt_misc rewriting interp to an ELF, recurses,
| loads ELF, happiness). Without the heap patch, we could be pointing
| into old stack (rewritten e.g. during module load or taking an
| interrupt, etc) on the loop retries. Without the ELOOP patch, the
| recursion could explode with an abusive script chain.

  I see! Thanks so much for explaining Kees, I appreciate it.

Thank you.
--
Prasad J Pandit / Red Hat Security Response Team
DB7A 84C5 D3F9 7CD1 B5EB  C939 D048 7860 3655 602B