On Sun, Jan 03, 2021 at 07:25:36PM +0100, Helge Deller wrote:
> On 1/3/21 3:27 PM, kernel test robot wrote:
> > FYI, we noticed the following commit (built with gcc-9):
> >
> > commit: 30a3a192730a997bc4afff5765254175b6fb64f3 ("[PATCH] proc/wchan: Use printk format instead of lookup_symbol_name()")
> > url: https://github.com/0day-ci/linux/commits/Helge-Deller/proc-wchan-Use-printk-format-instead-of-lookup_symbol_name/20201218-010048
> > base: https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git 09162bc32c880a791c6c0668ce0745cf7958f576
> >
> > in testcase: leaking-addresses
> > version: leaking-addresses-x86_64-4f19048-1_20201111
> > [...]
> > caused below changes (please refer to attached dmesg/kmsg for entire log/backtrace):
>
> I don't see anything wrong with the wchan patch (30a3a192730a997bc4afff5765254175b6fb64f3),
> or that it could have leaked anything.
>
> Maybe the kernel test robot picked up the wchan patch by mistake ?
>
> > [...]
> > [2 wchan] 0xffffc9000000003c
^^^^^
As the root cause of a kernel address exposure, Jann pointed out[2]
commit 152c432b128c, which I've tracked to here, only to discover this
regression was, indeed, reported. :(
So, we have a few things:
1) wchan has been reporting "0" in the default x86 config (ORC unwinder)
for 4 years now.
2) non-x86 or non-ORC, wchan has been leaking raw kernel addresses since
commit 152c432b128c (v5.12).
3) the output of scripts/leaking_addresses.pl is hard to read. :)
We can fix 1 and 2 with:
https://lore.kernel.org/lkml/20210923233105.4045080-1-keescook@xxxxxxxxxxxx/
(though that will need a Cc: stable now...)
If we don't do that, we still need to revert 152c432b128c in v5.12 and
later.
We should likely make leaking_addresses.pl a little more readable while
we're at it.
-Kees
[1] https://lore.kernel.org/lkml/20210921193249.el476vlhg5k6lfcq@xxxxxxxxxxxxxxxxxxxxxxxx/
[2] https://lore.kernel.org/lkml/CAG48ez2zC=+PuNgezH53HBPZ8CXU5H=vkWx7nJs60G8RXt3w0Q@xxxxxxxxxxxxxx/
--
Kees Cook
