[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <1b0ce523-f2a4-e525-b3b4-1034e77f69d6@linux.com>
Date: Wed, 15 Nov 2017 00:50:36 +0300
From: Alexander Popov <alex.popov@...ux.com>
To: Andy Lutomirski <luto@...nel.org>
Cc: "kernel-hardening@...ts.openwall.com"
<kernel-hardening@...ts.openwall.com>, Kees Cook <keescook@...omium.org>,
PaX Team <pageexec@...email.hu>, Brad Spengler <spender@...ecurity.net>,
Ingo Molnar <mingo@...nel.org>, Peter Zijlstra <peterz@...radead.org>,
Tycho Andersen <tycho@...ker.com>, Laura Abbott <labbott@...hat.com>,
Mark Rutland <mark.rutland@....com>,
Ard Biesheuvel <ard.biesheuvel@...aro.org>, Borislav Petkov <bp@...en8.de>,
Thomas Gleixner <tglx@...utronix.de>, "H . Peter Anvin" <hpa@...or.com>,
X86 ML <x86@...nel.org>
Subject: Re: [PATCH RFC v5 2/5] gcc-plugins: Add STACKLEAK plugin for tracking
the kernel stack
Hello Andy,
Thanks for your prompt reply!
On 14.11.2017 19:13, Andy Lutomirski wrote:
> On Tue, Nov 14, 2017 at 7:36 AM, Alexander Popov <alex.popov@...ux.com> wrote:
>> On 30.10.2017 21:06, Alexander Popov wrote:
>>> On 30.10.2017 20:32, Peter Zijlstra wrote:
>>>> On Mon, Oct 30, 2017 at 07:51:33PM +0300, Alexander Popov wrote:
>>>>> When the thread stack is exhausted, this BUG() is hit. But do_error_trap(),
>>>>> which handles the exception, calls track_stack() itself again (since it is
>>>>> instrumented by the gcc plugin). So this recursion proceeds with exhausting the
>>>>> thread stack.
>>>>
>>>> Add a __attribute__((nostacktrack)) on it?
>>>
>>> Yes, I already tried some blacklisting in the plugin, but it didn't really help,
>>> because:
>>>
>>> 1. there are other (more than 5) instrumented functions, that are called during
>>> BUG() handling too;
>>>
>>> 2. decreasing CONFIG_STACKLEAK_TRACK_MIN_SIZE would add more instrumented
>>> functions, which should be manually blacklisted (not good).
>>>
>>> I guess handling BUG() in another stack would be a solution. For example, Andy
>>> Lutomirski calls handle_stack_overflow in the DOUBLEFAULT_STACK
>>> (arch/x86/mm/fault.c). Should I do something similar?
>>
>> Hello Andy! May I ask your advice?
>>
>> When CONFIG_VMAP_STACK is disabled and STACKLEAK is enabled (for example, on
>> x86_32), we need another way to detect stack depth overflow. That is the reason
>> of having this BUG() in track_stack(). But it turns out to be recursive since
>> track_stack() will be called again during BUG() handling.
>
> What does the STEAKLACK plugin actually do? I haven't followed this enough.
I've just replied to Mark's explanation.
>> We can avoid that recursion by handling oops in another stack. It looks similar
>> to the way you call handle_stack_overflow() in arch/x86/mm/fault.c. But it seems
>> that I can't reuse that code, am I right?
>
> You'd probably have to make 32-bit compatible, which means making a
> 32-bit variant of this thingy:
>
> asm volatile ("movq %[stack], %%rsp\n\t"
> "call handle_stack_overflow\n\t"
> "1: jmp 1b"
> : ASM_CALL_CONSTRAINT
> : "D" ("kernel stack overflow (page fault)"),
> "S" (regs), "d" (address),
> [stack] "rm" (stack));
Hm, I don't have these pt_regs in track_stack(). That is why I think I can't
reuse your handle_stack_overflow(). I guess manually crafting pt_regs will not
be good-looking.
> Or you could force a double-fault.
Could you elaborate on that?
The goal is to have a verbose oops message and kill the offending process (if we
work on behalf of a process). Can I do that?
>> How should I do it properly?
>>
>> By the way, you wrote that you have some entry code changes which conflict with
>> STACKLEAK. May I ask for more details?
>
> It's this thing:
>
> https://git.kernel.org/pub/scm/linux/kernel/git/luto/linux.git/log/?h=x86/entry_stack.wip
>
> and I'll probably drop the ".wip" from the name shortly.
Wow, it's big. I'll look into it and maybe return with questions.
Best regards,
Alexander
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
