[lkml]   [2006]   [Jun]   [21]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
    SubjectRe: Possible bug in do_execve()
    On Wed, Jun 21, 2006 at 02:42:50PM -0500, Serge E. Hallyn wrote:
    > Quoting Sonny Rao (
    > > On Wed, Jun 21, 2006 at 02:09:10PM -0500, Serge E. Hallyn wrote:
    > > <snip>
    > > > > Yeah, I proposed a similar patch to Anton, and it would quiet the
    > > > > warning on powerpc, but that's not the point. It happens that powerpc
    > > > > doesn't use 0 as a context id, but that may not be true on another
    > > > > architecture. That's really what I'm concerned about.
    > > >
    > > > FWIW, ppc and cris do the NO_CONTEXT check, while others don't
    > > > even have a arch-specific 'mm->'.
    > >
    > > Good point. I probably stated that concern too narrowly. Probably
    > > what I should say is: What is the pre-condition for calling
    > > destroy_context() ? Is it that init_new_context() must have
    > > succeeded? Or is it merely that mm.context has been zeroed
    > > out?
    > Right, that may be the right question. If that's the case, then the
    > problem is really include/linux/sched.h:__mmdrop() which is what's
    > calling destroy_context(). Separating that out becomes a pretty
    > big patch affecting at least all mmput() and mmdrop() callers.

    So mmdrop() inlines to an atomic_dec_and_test on mm_count and a call
    to __mmdrop which makes three calls : mm_free_pgd(), destroy_context(),
    and free_mm(). I _think_ that in this case __mmdrop() will always get

    We know that the destroy_context() is unnecessary, but mm_free_pgd()
    and free_mm() are necessary.

    I was thinking we _could_ open code these calls in exec.c but that seems
    like a "Really Bad Idea" w.r.t abstraction/maintenance etc,
    and the alternative is to make another function/macro just for this
    special case, which also seems like a poor choice.

    > > It seems to assume that mm->context is valid before doing a check.
    > >
    > > Since I don't have a sparc64 box, I can't check to see if this
    > > actually breaks things or not.
    > So we can either go through all arch's and make sure destroy_context is
    > safe for invalid context, or split mmput() and destroy_context()...
    > The former seems easier, but the latter seems more robust in the face of
    > future code changes I guess.

    Yes, the former does seem easier, and perhaps easiest is to do that
    and document what the pre-conditions are so future developers at least
    have a clue.
    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to
    More majordomo info at
    Please read the FAQ at

     \ /
      Last update: 2006-06-21 22:15    [W:0.025 / U:40.284 seconds]
    ©2003-2016 Jasper Spaans. hosted at Digital OceanAdvertise on this site