[File] Indirect offset issue

[Christos Zoulas]

Fixed, thanks!

christos

> On Feb 13, 2019, at 5:59 PM, B Watson <yalhcru at gmail.com> wrote:
> 
> Working on some magic that uses indirect offsets, like:
> 
>>> (4.s-(-2))    ubyte    x
> 
> I found that this only works if there are at least 96 bytes in the
> file beyond the location of the offset. It turns out to be this, in
> softmagic.c, line 1531 (just before the switch):
> 
>         if (OFFSET_OOB(nbytes, offset + off, sizeof(*q)))
>            return 0;
> 
> sizeof(*q) is 96 on my platform, which explains the problem I was having.
> It looks like the intent is to check the size of whichever element in
> the union is the 'active' one, but instead it's getting the size of the
> largest element in the union.
> 
> The attached patch gets rid of this check, and instead adds OFFSET_OOB
> checks for the correct number of bytes at each switch label. It fixes the
> 96-byte problem and doesn't seem to break anything else, but I'm pretty
> new to file-hacking so I'd like to hear from someone more experienced.
> <file-softmagic-indirect-offset-fix_diff_gz.DEFANGED-9486><sanitizer.log>