Re: [PATCH 3/4] block: fix error return code

[Dan Carpenter <dan.carpenter@oracle.com>]

On Thu, Aug 07, 2014 at 05:48:24PM -0400, Jeff Moyer wrote:
> BTW, just above this there is questionable code:
> 
>         st = rsxx_get_num_targets(card, &card->n_targets);
>         if (st)
>                 dev_info(CARD_TO_DEV(card),
>                         "Failed reading the number of DMA targets\n");
> 
>         card->ctrl = kzalloc(card->n_targets * sizeof(*card->ctrl), GFP_KERNEL);
>         if (!card->ctrl) {
>                 st = -ENOMEM;
>                 goto failed_dma_setup;
>         }
> 
> >From my reading of the kzalloc code, ZERO_SIZE_PTR (which is 16 cast to
> a void *) would be returned from that kzalloc call if the
> rsxx_get_num_targets call failed (since you'd be kzalloc-ing 0 bytes).
> That would lead to the !card->ctrl check not working, right?
>

ZERO_SIZE_PTR is a subtle thing.  The if (!card->ctrl) check correctly
tells you if you allocated enough space to hold zero elements.  Which is
yes so we can continue without a problem.

Of course, you'd have to look at the surrounding code to see if there is
a problem...  I think I have seen dereferencing ZERO_SIZE_PTR bugs in
the past, but they are rare.

> I'd suggest not continuing after rsxx_get_num_targets fails.

Introducing new failures is a bad thing unless you know the code very
well or you can test it.  Your instinct should always be to not do that.

regards,
dan carpenter