Re: [BUG] ALSA: core: pcm_memory: a possible data race in do_alloc_pages()

From: Takashi Iwai
Date: Mon Jun 26 2023 - 03:09:07 EST


On Mon, 26 Jun 2023 05:42:29 +0200,
Tuo Li wrote:
>
>
> Hello,
>
> Our static analysis tool finds a possible data race in ALSA in Linux 6.4.0.
>
> In some functions, the field snd_card.total_pcm_alloc_bytes is accessed
> with holding the lock snd_card.memory_mutex. Here is an example:
>
>   do_free_pages() --> Line 57
>     mutex_lock(&card->memory_mutex); --> Line 61 (Lock card->memory_mutex)
>     card->total_pcm_alloc_bytes -= dmab->bytes;  --> Line 63 (Access  card->
> total_pcm_alloc_bytes)
>
> However, in the function do_alloc_pages():
>
>   if (max_alloc_per_card &&
>     card->total_pcm_alloc_bytes + size > max_alloc_per_card) --> Line 41
>
> the variable card->total_pcm_alloc_bytes is accessed without holding
> the lock card->memory_mutex, and thus a data race can occur.
>
> In my opinion, this data race may be harmful, because the value of
> card->total_pcm_alloc_bytes may be changed by another thread after
> the if check. Therefore, its value may be too large after Line 51 and can
> cause memory bugs such as buffer overflow:
>
>   card->total_pcm_alloc_bytes += dmab->bytes;  --> Line 51
>
> I am not quite sure whether this possible data race is real and how to
> fix it if it is real.
>
> Any feedback would be appreciated, thanks!
>
> Reported-by: BassCheck <bass@xxxxxxxxxxx>

It's a bit racy indeed, but the effect is almost negligible. The size
check there is merely a sanity check, and allocating more bytes
doesn't mean to conflict against anything practically.

That said, it's a better-to-be-addressed bug, but nothing too
serious.


thanks,

Takashi