linux-kernel - Re: [PATCH v2] vsprintf: protect kernel from panic due to non-canonical pointer dereference

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <288a7aba-c647-b51f-bdb3-99a62449dbb3@oracle.com>
Date:   Tue, 18 Oct 2022 19:36:41 +0000
From:   Jane Chu <jane.chu@...cle.com>
To:     Petr Mladek <pmladek@...e.com>
CC:     Andy Shevchenko <andriy.shevchenko@...ux.intel.com>,
        "rostedt@...dmis.org" <rostedt@...dmis.org>,
        "senozhatsky@...omium.org" <senozhatsky@...omium.org>,
        "linux@...musvillemoes.dk" <linux@...musvillemoes.dk>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        "linux-mm@...ck.org" <linux-mm@...ck.org>,
        Haakon Bugge <haakon.bugge@...cle.com>,
        John Haxby <john.haxby@...cle.com>,
        Jane Chu <jane.chu@...cle.com>
Subject: Re: [PATCH v2] vsprintf: protect kernel from panic due to
 non-canonical pointer dereference

On 10/18/2022 12:40 AM, Petr Mladek wrote:
> On Mon 2022-10-17 21:12:04, Jane Chu wrote:
>> On 10/17/2022 1:27 PM, Andy Shevchenko wrote:
>>> On Mon, Oct 17, 2022 at 01:44:47PM -0600, Jane Chu wrote:
>>>> While debugging a separate issue, it was found that an invalid string
>>>> pointer could very well contain a non-canical address, such as
>>>
>>> non-canical?
>>
>> Sorry, typo, will fix.
>>
>>>
>>>> 0x7665645f63616465. In that case, this line of defense isn't enough
>>>> to protect the kernel from crashing due to general protection fault
>>>>
>>>> 	if ((unsigned long)ptr < PAGE_SIZE || IS_ERR_VALUE(ptr))
>>>>                   return "(efault)";
>>>>
>>>> So run one more round of check via kern_addr_valid(). On architectures
>>>> that provide meaningful implementation, this line of check effectively
>>>> catches non-canonical pointers, etc.
>>>
>>> OK, but I don't see how this is useful in the form of returning efault here.
>>> Ideally we should inform user that the pointer is wrong and how it's wrong.
>>> But. It will crash somewhere else at some point, right?
>> Broadly speaking, yes.  It's not a perfect line of defense, but again,
>> the bug scenario is a "cat" of some sysfs attributes that leads to
>> panic. Does it make sense for kernel to protect itself against panic
>> triggered by a "cat" from user if it could?
> 
>  From my view, the check might be useful. I agree with Andy that the
> broken pointer would cause crash later anyway. But the check
> in vsprintf() would allow to see a message that the pointer was
> wrong before the system crashes.
> 
> That said, this was much more important in the past when printk()
> called vsprintf() under logbuf_lock. Nested printk() calls
> were redirected to per-CPU buffers and eventually lost.
> 
> It works better now when printk() uses lockless ringbuffer and
> vsprintf() is called twice there. The first call is used
> to get the string length so that it could reserve the needed
> space in the ring buffer. If vsprintf() crashes during the 1st call
> then it would be possible to print the nested Oops messages.
> 
> 
>> I mean that there
>>> is no guarantee that kernel has protection in every single place against
>>> dangling / invalid pointers. One way or another it will crash.
>>>
>>> That said, honestly I have no idea how this patch may be considered
>>> anything but band-aid. OTOH, I don't see a harm. Perhaps others will
>>> share their opinions.
>>>
>>
>> 3+ years ago, commit 3e5903eb9cff7 ("vsprintf: Prevent crash when
>> dereferencing invalid pointers") provided the similar level of
>> protection as this patch.  But it was soon revised by commit
>> 2ac5a3bf7042a ("vsprintf: Do not break early boot with probing
>> addresses"), and that's why the string() utility no longer detects
>> non-canonical string pointer.
>>
>> I only thought that kern_addr_valid() is less of a heavy hammer, and
>> could be safely deployed.
> 
> Hmm, I see that kern_addr_valid() is available (defined) only on some
> architectures. This patch would break build on architectures where it
> is not defined.
> 
> Also it is used only when reading /proc/kcore. It means that it is not
> tested during early boot. I wonder if it actually works during
> the very early boot.
> 
> Result:
> 
> The patch is not usable as is.
> 
> IMHO, it is not worth the effort to get it working. Especially when
> printk() should be able to show Oops inside vsprintf() these days.

Yes, kern_addr_valid() is used by read_kcore() which is architecturally 
independent and applies everywhere, so does that imply that it is 
defined in all architectures?

I guess the early boot scenario is different in that, potentially unkind 
users aren't involved, hence a broken kernel is broken and need a fix.

The scenario concerned here is with users could potentially exploit a 
kernel issue with DOS attack.  Then we have the scenario that the kernel
bug itself is confined, in that, had the sysfs not been accessed, the 
OOB pointer won't be produced.  So in this case, "(efault)" is a lot 
more desirable than panic.

> 
> Regards,
> Petr

Thanks!
-jane