Received-SPF: pass (google.com: domain of linux-kernel+bounces-178499-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) client-ip=2604:1380:45e3:2400::1;
From: Thomas Gleixner <tglx@linutronix.de>
To: Justin Stitt <justinstitt@google.com>, John Stultz <jstultz@google.com>,
 Stephen Boyd <sboyd@kernel.org>, Nathan Chancellor <nathan@kernel.org>,
 Bill Wendling <morbo@google.com>
Cc: linux-kernel@vger.kernel.org, llvm@lists.linux.dev, Justin Stitt
 <justinstitt@google.com>
Subject: Re: [PATCH] ntp: safeguard against time_constant overflow case
In-Reply-To: <20240506-b4-sio-ntp-c-v1-1-a01281aa01ba@google.com>
References: <20240506-b4-sio-ntp-c-v1-1-a01281aa01ba@google.com>
Date: Tue, 14 May 2024 11:17:23 +0200
Message-ID: <87y18clplo.ffs@tglx>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain

On Mon, May 06 2024 at 22:01, Justin Stitt wrote:

> Using syzkaller with the recently reintroduced signed integer overflow
> sanitizer produces this UBSAN report:
>
> [   46.809326] ------------[ cut here ]------------
> [   46.812882] UBSAN: signed-integer-overflow in ../kernel/time/ntp.c:738:18
> [   46.817676] 9223372036854775806 + 4 cannot be represented in type 'long'
> [   46.822346] CPU: 1 PID: 685 Comm: syz-executor.0 Not tainted 6.8.0-rc2-00036-g679ee73ec453 #2
> [   46.828270] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.16.3-debian-1.16.3-2 04/01/2014
> [   46.834836] Call Trace:
> [   46.836625]  <TASK>
> [   46.838147]  dump_stack_lvl+0x93/0xd0
> [   46.840771]  handle_overflow+0x171/0x1b0
> [   46.843516]  __do_adjtimex+0x1236/0x1440
> [   46.846275]  do_adjtimex+0x2be/0x740
> [   46.848864]  __x64_sys_clock_adjtime+0x154/0x1d0
> [   46.852164]  do_syscall_64+0xd7/0x1b0
> [   46.854783]  ? arch_exit_to_user_mode_prepare+0x11/0x60
> [   46.858426]  entry_SYSCALL_64_after_hwframe+0x6f/0x77
> [   46.861914] RIP: 0033:0x7fde90aaf539
> [   46.864500] Code: 28 00 00 00 75 05 48 83 c4 28 c3 e8 f1 14 00 00 90 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 c7 c1 8
> [   46.877151] RSP: 002b:00007ffebfe63358 EFLAGS: 00000246 ORIG_RAX: 0000000000000131
> [   46.882279] RAX: ffffffffffffffda RBX: 00007fde90be3f80 RCX: 00007fde90aaf539
> [   46.887270] RDX: 0000000000000000 RSI: 0000000020000280 RDI: 0000000000000000
> [   46.892174] RBP: 00007fde90b0e496 R08: 0000000000000000 R09: 0000000000000000
> [   46.897061] R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000000
> [   46.902020] R13: 0000000000000095 R14: 00007fde90be3f80 R15: 00007fde90be3f80
> [   46.906946]  </TASK>
> [   46.908537] ---[ end trace ]---

Please trim stack traces so they contain only useful information.

 UBSAN: signed-integer-overflow in ../kernel/time/ntp.c:738:18
 9223372036854775806 + 4 cannot be represented in type 'long'
 Call Trace:
  <TASK>
  handle_overflow+0x171/0x1b0
  __do_adjtimex+0x1236/0x1440
  do_adjtimex+0x2be/0x740
  __x64_sys_clock_adjtime+0x154/0x1d0
  do_syscall_64+0xd7/0x1b0

Is completely sufficient, no?

> Historically, the signed integer overflow sanitizer did not work in the
> kernel due to its interaction with `-fwrapv` but this has since been
> changed [1] in the newest version of Clang; It being re-enabled in the
> kernel with Commit 557f8c582a9ba8ab ("ubsan: Reintroduce signed overflow
> sanitizer").

How is that relevant to the problem?

> Nonetheless, let's slightly rework the logic surrounding time_constant

s/Nonetheless, let's slightly /Rework/

> and how it is incremented such that we avoid unintentional wrap-around
> (even though it is extremely unlikely to be hit in non-fuzzing
> scenarios).

We don't avoid anything. Please write change logs in imperative mood.

>  	if (txc->modes & ADJ_TIMECONST) {
>  		time_constant = txc->constant;
> -		if (!(time_status & STA_NANO))
> -			time_constant += 4;
> -		time_constant = min(time_constant, (long)MAXTC);
> -		time_constant = max(time_constant, 0l);
> +		if (!(time_status & STA_NANO) &&
> +		    unlikely(LONG_MAX - time_constant_inc >= time_constant))

What's unlikely about this? Correct operation of adjtimex() will
increment, no?

As this obviously will be clamped to MAXTC anyway, you can spare that whole
LONG_MAX - time_constant_inc dance and simply do:

		if (!(time_status & STA_NANO) && time_constant < MAXTC)
                	time_constant += 4;

No?

> +			time_constant += time_constant_inc;
> +		time_constant = clamp_t(long, time_constant, 0, MAXTC);

Thanks,

        tglx