Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
References: <20200317132157.1272-1-cai@lca.pw>
In-Reply-To: <20200317132157.1272-1-cai@lca.pw>
From:   Marco Elver <elver@google.com>
Date:   Tue, 17 Mar 2020 14:31:18 +0100
Message-ID: <CANpmjNPRTEsaTxDeMnz8T6BtfEx5yzgFCYNP2KZkedZ2kq9ZUw@mail.gmail.com>
Subject: Re: [PATCH -next] mm/kmemleak: annotate a data race in checksum
To:     Qian Cai <cai@lca.pw>
Cc:     Andrew Morton <akpm@linux-foundation.org>, catalin.marinas@arm.com,
        Linux Memory Management List <linux-mm@kvack.org>,
        LKML <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Tue, 17 Mar 2020 at 14:22, Qian Cai <cai@lca.pw> wrote:
>
> Even if KCSAN is disabled for kmemleak, update_checksum() could still
> call crc32() (which is outside of kmemleak.c) to dereference
> object->pointer. Thus, the value of object->pointer could be accessed
> concurrently as noticed by KCSAN,
>
>  BUG: KCSAN: data-race in crc32_le_base / do_raw_spin_lock
>
>  write to 0xffffb0ea683a7d50 of 4 bytes by task 23575 on cpu 12:
>   do_raw_spin_lock+0x114/0x200
>   debug_spin_lock_after at kernel/locking/spinlock_debug.c:91
>   (inlined by) do_raw_spin_lock at kernel/locking/spinlock_debug.c:115
>   _raw_spin_lock+0x40/0x50
>   __handle_mm_fault+0xa9e/0xd00
>   handle_mm_fault+0xfc/0x2f0
>   do_page_fault+0x263/0x6f9
>   page_fault+0x34/0x40
>
>  read to 0xffffb0ea683a7d50 of 4 bytes by task 839 on cpu 60:
>   crc32_le_base+0x67/0x350
>   crc32_le_base+0x67/0x350:
>   crc32_body at lib/crc32.c:106
>   (inlined by) crc32_le_generic at lib/crc32.c:179
>   (inlined by) crc32_le at lib/crc32.c:197
>   kmemleak_scan+0x528/0xd90
>   update_checksum at mm/kmemleak.c:1172
>   (inlined by) kmemleak_scan at mm/kmemleak.c:1497
>   kmemleak_scan_thread+0xcc/0xfa
>   kthread+0x1e0/0x200
>   ret_from_fork+0x27/0x50
>
> If a shattered value was returned due to a data race, it will be
> corrected in the next scan. Thus, annotate it as an intentional data
> race using the data_race() macro.
>
> Signed-off-by: Qian Cai <cai@lca.pw>
> ---
>  mm/kmemleak.c | 7 ++++++-
>  1 file changed, 6 insertions(+), 1 deletion(-)
>
> diff --git a/mm/kmemleak.c b/mm/kmemleak.c
> index e362dc3d2028..d3327756c3a4 100644
> --- a/mm/kmemleak.c
> +++ b/mm/kmemleak.c
> @@ -1169,7 +1169,12 @@ static bool update_checksum(struct kmemleak_object *object)
>         u32 old_csum = object->checksum;
>
>         kasan_disable_current();

Suggested:
+ kcsan_disable_current();

> -       object->checksum = crc32(0, (void *)object->pointer, object->size);
> +       /*
> +        * crc32() will dereference object->pointer. If an unstable value was
> +        * returned due to a data race, it will be corrected in the next scan.
> +        */
> +       object->checksum = data_race(crc32(0, (void *)object->pointer,
> +                                          object->size));

This will work with the default config, because for word-sized-aligned
writes no marking is enforced. But this will still cause a data race
if the write is e.g. due to a memcpy.

There are already markers for KASAN around, so the most reliable thing
is to just disable KCSAN in this region.

>         kasan_enable_current();

Suggested:
+ kcsan_enable_current();

Thanks,
-- Marco

>         return object->checksum != old_csum;
> --
> 2.21.0 (Apple Git-122.2)
>