Re: [PATCH] tmpfs: fix uninitialized return value in shmem_link

[Hugh Dickins]

On Mon, 25 Feb 2019, Linus Torvalds wrote:
> On Fri, Feb 22, 2019 at 10:35 PM Hugh Dickins <hughd@xxxxxxxxxx> wrote:
> >
> > When we made the shmem_reserve_inode call in shmem_link conditional, we
> > forgot to update the declaration for ret so that it always has a known
> > value.  Dan Carpenter pointed out this deficiency in the original patch.
> 
> Applied.

Thanks.  And I apologize for letting that slip through: Darrick sent
the patch fragment, I dressed it up, and more or less tricked him into
taking ownership of the bug, when it's I who should have been more careful.

But I'm glad it confirmed your rc8 instinct, rather than messing final :)

> 
> Side note: how come gcc didn't warn about this? Yes, we disable that
> warning for some cases because of lots of false positives, but I
> thought the *default* setup still had it.

I thought so too, and have been puzzled by it.  If I try removing the
initialization of inode from the next function, shmem_unlink(), I do
get the expected warning for that.

> 
> Is it just that the goto ends up confusing gcc enough that it never notices?

Since the goto route did have ret properly initialized, I don't see
why it might have been confusing, but what do I know...

I thought it might be because outside the goto route, ret was used
for nothing but the return value.  But that's disproved: I tried a
very silly "inode->i_flags = ret;" just after d_instantiate(),
and still no warning when ret is uninitialized.

Seems like a gcc bug? But I don't have a decent recent gcc to hand
to submit a proper report, hope someone else can shed light on it.

Hugh