Re: [PATCH] xfs: fix possible NULL dereference

[Dave Chinner <david@fromorbit.com>]

On Tue, Oct 22, 2013 at 04:19:44PM -0500, Eric Sandeen wrote:
> On 10/22/13 4:03 PM, Dave Chinner wrote:
> > On Tue, Oct 22, 2013 at 03:49:01PM -0500, Eric Sandeen wrote:
> >> On 10/22/13 3:39 PM, Dave Chinner wrote:
> >>> On Tue, Oct 22, 2013 at 08:12:51AM -0200, Geyslan Gregório Bem wrote:
> >>>> 2013/10/21 Dave Chinner <david@fromorbit.com>:
> >>>>> On Mon, Oct 21, 2013 at 07:00:59PM -0500, Eric Sandeen wrote:
> >>>>>> On 10/21/13 6:56 PM, Dave Chinner wrote:
> >>>>>>> On Mon, Oct 21, 2013 at 06:18:49PM -0500, Ben Myers wrote:
> >>>>>
> >>>>> Yes, but to continue the Devil's Advocate argument, the purpose of
> >>>>> debug code isn't to enlightent the casual reader or drive-by
> >>>>> patchers - it's to make life easier for people who actually spend
> >>>>> time debugging the code. And the people who need the debug code
> >>>>> are expected to understand why an ASSERT is not necessary. :)
> >>>>>
> >>>> Dave, Eric and Ben,
> >>>>
> >>>> This was catched by coverity (CID 102348).
> >>>
> >>> You should have put that in the patch description.
> >>>
> >>> Now I understand why there's been a sudden surge of irrelevant one
> >>> line changes from random people that have never touched XFS before.
> >>>
> >>> <sigh>
> >>>
> >>> Ok, lets churn the code just to shut the stupid checker up. This
> >>> doesn't fix a bug, it doesn't change behaviour, it just makes
> >>> coverity happy. Convert it to the for loop plus ASSERT I mentioned
> >>> in a previous message.
> >>
> >> You know, I respectfully disagree, but we might just have to agree
> >> to disagree.  The code, as it stands, tests for a null ptr
> >> and then dereferences it.  That's always going to raise some
> >> eyebrows, coverity or not, debug code or not, drive by or not.
> > 
> >> So even for future developers, making the code more self-
> >> documenting about this behavior would be a plus, whether it's by
> >> comment, by explicit ASSERT(), or whatever.  (I don't think
> >> that xfs_emerg() has quite enough context to make it obvious.)
> > 
> > Sure, but if weren't for the fact that Coverity warned about it,
> > nobody other that us people who work on the XFS code day in, day out
> > would have even cared about it.
> > 
> > That's kind of my point - again, as the Devil's Advocate - that
> > coverity is encouraging drive-by "fixes" by people who don't
> > actually understand any of the context, history and/or culture
> > surrounding the code being modified.
> 
> They shouldn't have to, the code (or comments therein) should
> make it obvious.  ;)  (in a perfect world...)

Obvious to whom, exactly?

That's the point I'm trying to make - "#ifdef DEBUG", two
comments indicating that it's validating the list and printing a
message just before it goes boom. That's pretty obvious code to
anyone who is used to tracking down corrupted list problems...

> > I have no problems with real bugs being fixed, but if we are
> > modifying code for no gain other than closing "coverity doesn't like
> > it" bugs, then we *should* be questioning whether the change is
> > really necessary.
> 
> But let's give Geyslan the benefit of the doubt, and realize that
> Coverity does find real things, and even if it originated w/ a
> Coverity CID, when one sees:
> 
> 	if (!a)
> 		printk("a thing\n")
> 
> 	a = a->b = . . . 
> 
> it looks suspicious to pretty much anyone.  I don't think Geyslan
> sent it to shut Coverity up, he sent it because it looked like
> a bug worth fixing (after Coverity spotted it).
> 
> Let's not be too hard on him for trying; I appreciate it more
> than spelling fixes and whitespace cleanups.  ;)

True, point taken. 

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com