linux-kernel - Re: [PATCH 0/6] Allow signals for IO threads

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] [day] [month] [year] [list]

Message-ID: <0375b37f-2e1e-7999-53b8-c567422aa181@samba.org>
Date:   Sat, 3 Apr 2021 02:48:26 +0200
From:   Stefan Metzmacher <metze@...ba.org>
To:     Linus Torvalds <torvalds@...ux-foundation.org>
Cc:     Jens Axboe <axboe@...nel.dk>, io-uring <io-uring@...r.kernel.org>,
        "Eric W. Biederman" <ebiederm@...ssion.com>,
        Oleg Nesterov <oleg@...hat.com>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH 0/6] Allow signals for IO threads

Am 01.04.21 um 18:24 schrieb Linus Torvalds:
> On Thu, Apr 1, 2021 at 9:00 AM Stefan Metzmacher <metze@...ba.org> wrote:
>>
>> I haven't tried it, but it seems gdb tries to use PTRACE_PEEKUSR
>> against the last thread tid listed under /proc/<pid>/tasks/ in order to
>> get the architecture for the userspace application
> 
> Christ, what an odd hack. Why wouldn't it just do it on the initial
> thread you actually attached to?
> 
> Are you sure it's not simply because your test-case was to attach to
> the io_uring thread? Because the io_uring thread might as well be
> considered to be 64-bit.

      │   └─io_uring-cp,1396 Makefile file
      │       ├─{iou-mgr-1396},1397
      │       ├─{iou-wrk-1396},1398
      │       └─{iou-wrk-1396},1399

strace -ttT -o strace-uring-fail.txt gdb --pid 1396
(note strace -f would deadlock gdb with SIGSTOP)

The full file can be found here:
https://www.samba.org/~metze/strace-uring-fail.txt
(I guess there was a race and the workers 1398 and 1399 exited in between,
that's it using 1397):

18:46:35.429498 ptrace(PTRACE_PEEKUSER, 1397, 8*CS, [NULL]) = 0 <0.000022>

>> so my naive assumption
>> would be that it wouldn't allow the detection of a 32-bit application
>> using a 64-bit kernel.
> 
> I'm not entirely convinced we want to care about a confused gdb
> implementation and somebody debugging a case that I don't believe
> happens in practice.
> 
> 32-bit user space is legacy. And legacy isn't io_uring. If somebody
> insists on doing odd things, they can live with the odd results.

Ok, I'd agree for 32-bit applications, but what about libraries?
E.g. distributions ship libraries like libsmbclient or nss modules
as 64-bit and 32-bit version, in order to support legacy applications
to run. Why shouldn't the 32-bit library builds not support io_uring?
(Note libsmbclient don't use io_uring yet, but I guess it will be in future).

Any ideas regarding similar problems for other architectures?

metze