Re: two bogus patches arising from CVE-2019-12381

[Florian Westphal <fw@strlen.de>]

Charles Bryant <ch.4g7vxy-nbkl8p@chch.co.uk> wrote:
> I believe two patches from last year are mistaken. They are:
> 
> https://git.kernel.org/pub/scm/linux/kernel/git/netdev/net.git/commit/?id=95baa60a0da80a0143e3ddd4d3725758b4513825
> 
> https://git.kernel.org/pub/scm/linux/kernel/git/netdev/net.git/commit/?id=425aa0e1d01513437668fa3d4a971168bbaa8515
> 
> Both of these make a function return immediately with ENOMEM if a kalloc()
> fails.  However in each case the function already correctly handled
> allocation failure later on. Furthermore, by making them exit early
> on allocation failure, it (very slightly) makes them worse as in some
> cases they might have correctly returned EADDRINUSE and not needed the
> allocated memory.
> 
> I think, therefore, that these changes should be reverted.

Both fixes are useless, as you explained above.

But they do not matter.  When GFP_KERNEL allocations fail the entire
system is screwed anyway.

So instead of revert, I would suggest that you wait until net-next
reopens, then:

For the first commit, send a patch that reverts, but also add a comment
that explains the error is handled below the loop.

For the second commit, send a patch that moves the allocation to where its
needed -- the spinlock was converted to a mutex so there is no need
for this ahead-of-time allocation anymore.