Subject: Re: [RFC 3/3] x86/ptrace: down with test_thread_flag(TIF_IA32)
To: Andy Lutomirski <luto@amacapital.net>, Oleg Nesterov <oleg@redhat.com>
References: <1461600763-3534-1-git-send-email-dsafonov@virtuozzo.com>
 <1461600763-3534-3-git-send-email-dsafonov@virtuozzo.com>
 <CALCETrU5O66hT0fwRLP-hwJ9_CK1Q4F2vSXRXZPQ1Je7Y1CKoQ@mail.gmail.com>
 <571E5076.2040802@virtuozzo.com> <571E5D51.2060809@virtuozzo.com>
CC: "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
        Dmitry Safonov <0x7f454c46@gmail.com>,
        Andy Lutomirski <luto@kernel.org>, Ingo Molnar <mingo@redhat.com>,
        Thomas Gleixner <tglx@linutronix.de>, "H. Peter Anvin" <hpa@zytor.com>,
        X86 ML <x86@kernel.org>
From: Dmitry Safonov <dsafonov@virtuozzo.com>
Message-ID: <571E5E3B.7020806@virtuozzo.com>
Date: Mon, 25 Apr 2016 21:13:15 +0300
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:38.0) Gecko/20100101
 Thunderbird/38.7.1
MIME-Version: 1.0
In-Reply-To: <571E5D51.2060809@virtuozzo.com>
Content-Type: text/plain; charset="utf-8"; format=flowed
Content-Transfer-Encoding: 7bit
SpamDiagnosticOutput: 1:23
SpamDiagnosticMetadata: NSPM
X-MS-Exchange-CrossTenant-OriginalArrivalTime: 25 Apr 2016 18:14:12.7617 (UTC)
X-MS-Exchange-CrossTenant-FromEntityHeader: Hosted
X-MS-Exchange-Transport-CrossTenantHeadersStamped: DB6PR0801MB1302
Sender: linux-kernel-owner@vger.kernel.org
List-ID: <linux-kernel.vger.kernel.org>
X-Mailing-List: linux-kernel@vger.kernel.org
Content-Length: 2928
Lines: 83

On 04/25/2016 09:09 PM, Dmitry Safonov wrote:
> On 04/25/2016 08:14 PM, Dmitry Safonov wrote:
>> On 04/25/2016 07:53 PM, Andy Lutomirski wrote:
>>> On Mon, Apr 25, 2016 at 9:12 AM, Dmitry Safonov 
>>> <dsafonov@virtuozzo.com> wrote:
>>>> As the task isn't executing at the moment of {GET,SET}REGS,
>>>> return regset that corresponds to code selector.
>>>> So, for i386 elf binary that changed it's CS to __USER_CS
>>>> it will return full x86_64 register set.
>>>>
>>>> That will change ABI: i.e, strace uses returned register size
>>>> to determine, in which mode the application is.
>>>> With the current ABI that way is buggy:
>>> Oleg, any comment here?
>>>
>>>> int main(int argc, char **argv, char **envp)
>>>> {
>>>>          printf("Here we exit\n");
>>>>          fflush(stdout);
>>>>          asm volatile ("int $0x80" : : "a" (__NR_exit), "D" (1));
>>>>          printf("After exit\n");
>>>>
>>>>          return 0;
>>>> }
>>>>
>>>> This program will confuse strace:
>>>>
>>>> [tst]$ strace ./confuse 2>&1 | tail
>>>> brk(0x1ca1000)                          = 0x1ca1000
>>>> write(1, "Here we exit\n", 13Here we exit
>>>> )          = 13
>>>> exit(1)                                 = ?
>>>> <... exit resumed> strace: _exit returned!
>>>> )                    = ?
>>>> write(1, "After exit\n", 11After exit
>>>> )            = 11
>>>> exit_group(0)                           = ?
>>>> +++ exited with 0 +++
>>>>
>>>> So this ABI change should make PTRACE_GETREGSET more reliable and
>>>> this will be another step to drop TIF_{IA32,X32} flags.
>>> Does strace start working again with this change?  I suspect that
>>> we'll eventually have to expose syscall_get_arch directly through
>>> ptrace, but that's a project for another day.
>>
>> Oh, crap, not yet - seems like, I failed with my test.
>> I'll resend this patch as will get it fixed, sorry.
>
> I find out, what I have changed (and broke test):
> > if (!user_64bit_mode(task_pt_regs(task)))
> was
> > if (task_thread_info(task)->status & TS_COMPAT)
>
> That way the test runs now:
>> brk(NULL)                               = 0x1145000
>> brk(0x1167000)                          = 0x1167000
>> write(1, "Here we exit\n", 13Here we exit
>> )          = 13
>> strace: [ Process PID=1608 runs in 32 bit mode. ]
>> umask(0)                                = 022

And that seems to be right as __NR_exit for amd64
is 60, which is the same as __NR_umask for i386.

>> strace: [ Process PID=1608 runs in 64 bit mode. ]
>> write(1, "After exit\n", 11After exit
>> )            = 11
>> exit_group(0)                           = ?
>> +++ exited with 0 +++
>
> But I changed on signal patch rebase and now I'm
> thinking: should it be
> > if (task_thread_info(task)->status & TS_COMPAT || 
> !user_64bit_mode(task_pt_regs(task)))
> or what?
> Should we count program that does compat syscall
> as compatible even if it's in 64-bit mode?
>


-- 
Regards,
Dmitry Safonov