[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CAGXu5jJ_8_Q_HukwuBEdtK4NQGp-Fnxhd2-1hXu7-KBfGX8CBw@mail.gmail.com>
Date: Wed, 26 Oct 2016 13:44:50 -0700
From: Kees Cook <keescook@...omium.org>
To: "Reshetova, Elena" <elena.reshetova@...el.com>
Cc: AKASHI Takahiro <takahiro.akashi@...aro.org>,
"kernel-hardening@...ts.openwall.com" <kernel-hardening@...ts.openwall.com>,
Hans Liljestrand <ishkamiel@...il.com>, David Windsor <dwindsor@...il.com>
Subject: Re: [RFC v2 PATCH 01/13] Add architecture
independent hardened atomic base
On Wed, Oct 26, 2016 at 3:27 AM, Reshetova, Elena
<elena.reshetova@...el.com> wrote:
> On Tue, Oct 25, 2016 at 11:20 AM, Reshetova, Elena <elena.reshetova@...el.com> wrote:
>>> struct list_head {
>>> diff --git a/kernel/panic.c b/kernel/panic.c index e6480e2..cb1d6db
>>> 100644
>>> --- a/kernel/panic.c
>>> +++ b/kernel/panic.c
>>> @@ -616,3 +616,14 @@ static int __init oops_setup(char *s)
>>> return 0;
>>> }
>>> early_param("oops", oops_setup);
>>> +
>>> +#ifdef CONFIG_HARDENED_ATOMIC
>>> +void hardened_atomic_overflow(struct pt_regs *regs) {
>>> + pr_emerg(KERN_EMERG "HARDENED_ATOMIC: overflow detected in: %s:%d, uid/euid: %u/%u\n",
>>> + current->comm, task_pid_nr(current),
>>> + from_kuid_munged(&init_user_ns, current_uid()),
>>> + from_kuid_munged(&init_user_ns, current_euid()));
>>> + BUG();
>>
>> BUG() will print a message like "kernel BUG at kernel/panic.c:627!"
>> and a stack trace dump with extra frames including hardened_atomic_overflow() and some exception handler routines (do_trap() on x86), which are totally useless. So I don't want to call BUG() here.
>>
>> Instead, we will fall back to a normal "BUG" handler, bug_handler() on arm64, which eventually calls die(), generating more *intuitive* messages:
>> ===8<===
>> [ 29.082336] lkdtm: attempting good atomic_add_return
>> [ 29.082391] lkdtm: attempting bad atomic_add_return
>> [ 29.082830] ------------[ cut here ]------------
>> [ 29.082889] Kernel BUG at ffff0000008b07fc [verbose debug info unavailable]
>> (Actually, this is lkdtm_ATOMIC_ADD_RETURN_OVERFLOW)
>> [ 29.082968] HARDENED_ATOMIC: overflow detected in: insmod:1152, uid/euid: 0/0
>> [ 29.083043] Internal error: Oops - BUG: 0 [#1] PREEMPT SMP
>> [ 29.083098] Modules linked in: lkdtm(+)
>> [ 29.083189] CPU: 1 PID: 1152 Comm: insmod Not tainted 4.9.0-rc1-00024-gb757839-dirty #12
>> [ 29.083262] Hardware name: FVP Base (DT)
>> [ 29.083324] task: ffff80087aa21900 task.stack: ffff80087a36c000
>> [ 29.083557] PC is at lkdtm_ATOMIC_ADD_RETURN_OVERFLOW+0x6c/0xa0 [lkdtm]
>> [ 29.083627] LR is at 0x7fffffff
>> [ 29.083687] pc : [<ffff0000008b07fc>] lr : [<000000007fffffff>] pstate: 90400149
>> [ 29.083757] sp : ffff80087a36fbe0
>> [ 29.083810] x29: ffff80087a36fbe0 [ 29.083858] x28: ffff000008ec3000
>> [ 29.083906]
>>
>> ...
>>
>> [ 29.090842] [<ffff0000008b07fc>] lkdtm_ATOMIC_ADD_RETURN_OVERFLOW+0x6c/0xa0 [lkdtm]
>> [ 29.091090] [<ffff0000008b20a4>] lkdtm_do_action+0x1c/0x28 [lkdtm]
>> [ 29.091334] [<ffff0000008bb118>] lkdtm_module_init+0x118/0x210 [lkdtm]
>> [ 29.091422] [<ffff000008083150>] do_one_initcall+0x38/0x128
>> [ 29.091503] [<ffff000008166ad4>] do_init_module+0x5c/0x1c8
>> [ 29.091586] [<ffff00000812e1ec>] load_module+0x1b24/0x20b0
>> [ 29.091670] [<ffff00000812e920>] SyS_init_module+0x1a8/0x1d8
>> [ 29.091753] [<ffff000008082ef0>] el0_svc_naked+0x24/0x28
>> [ 29.091843] Code: 910063a1 b8e0003e 2b1e0010 540000c7 (d4210020)
>> ===>8===
>>
>> So, you propose to remove call to BUG() fully from there? Funny, I think on x86 the output was actually like you wanted with just calling BUG().
>
> The x86 BUG isn't as nice:
> - "kernel BUG at kernel/panic.c:627" is bogus, the bug is a frame above, etc
> - the meaningful message "HARDENED_ATOMIC: overflow detected" happens above the ==cut== line
>
> Ok, what should we use instead then? Should I go back to the previous version and print this in addition:
>
> print_symbol(KERN_EMERG "HARDENED_ATOMIC: refcount overflow occurred at: %s\n", instruction_pointer(regs));
For now, we can stick to BUG(), but we'll find a way to improve it in
the future. I'll want to change these for HARDENED_ATOMIC,
HARDENED_USERCOPY, and BUG_ON_CORRUPTION (in -next), so it's not
specific to this series.
I'm open to Takahiro's suggestions for how to actually make these
changes, though. Notably neither bug_handler() nor die() are exported
outside the respective arch/ trees, so it's not clear what needs
changing. But it'll likely be separate from this series. :)
-Kees
--
Kees Cook
Nexus Security
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.
