Re: [nft PATCH 3/4] segtree: Fix for potential NULL-pointer deref in ei_insert()

[Phil Sutter <phil@nwl.cc>]

On Mon, Jan 27, 2020 at 02:24:48PM +0100, Phil Sutter wrote:
> Hi Pablo,
> 
> On Tue, Jan 21, 2020 at 01:56:12PM +0100, Pablo Neira Ayuso wrote:
> > On Mon, Jan 20, 2020 at 05:25:39PM +0100, Phil Sutter wrote:
> > > Covscan complained about potential deref of NULL 'lei' pointer,
> > > Interestingly this can't happen as the relevant goto leading to that
> > > (in line 260) sits in code checking conflicts between new intervals and
> > > since those are sorted upon insertion, only the lower boundary may
> > > conflict (or both, but that's covered before).
> > > 
> > > Given the needed investigation to proof covscan wrong and the actually
> > > wrong (but impossible) code, better fix this as if element ordering was
> > > arbitrary to avoid surprises if at some point it really becomes that.
> > > 
> > > Fixes: 4d6ad0f310d6c ("segtree: check for overlapping elements at insertion")
> > 
> > Not fixing anything. Tell them to fix covscan :-)
> 
> Well, I guess covscan is simply not intelligent enough to detect the
> impact of previous element sorting. :)

Or maybe I am not intelligent enough to read and comprehend the sorting
function. ;)

Meanwhile I managed to find a reproducer for covscan's complaint:

With a ruleset of:

| table ip t {
| 	set s {
| 		type inet_service
| 		flags interval
| 	}
| }

The following command segfaults:

| # nft add element t s '{ 10-40, 5-15 }'

According to gdb it happens the line above
'return expr_binary_error(...)' in ei_insert(), namely segtree.c:279. No
idea why it's in the wrong line, but it seems to be just the reported
issue as 'lei' is NULL.

Interestingly, adding a rule with an anonymous set and the same elements
works fine, no idea why.

What do you think, should I continue investigating or can we just go
with my original fix (for now at least)?

Cheers, Phil