[PATCH/RFC 0/2] ALSA: firewire: Fix integer overflows on 32-bit

[PATCH/RFC 0/2] ALSA: firewire: Fix integer overflows on 32-bit

[Geert Uytterhoeven]

	Hi all,

This patch series fixes two integer overflows on 32-bit platforms when
in multiplications with the NSEC_PER_SEC constant, found by code
inspection.  They are marked "RFC", as I don't know the maximum transfer
length of MIDI.  Depending on this maximum length, a small tweak may be
necessary.

Thanks for your comments!

Geert Uytterhoeven (2):
  ALSA: fireface: Fix integer overflow in transmit_midi_msg()
  ALSA: firewire-tascam: Fix integer overflow in midi_port_work()

 sound/firewire/fireface/ff-transaction.c   | 2 +-
 sound/firewire/tascam/tascam-transaction.c | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

-- 
2.25.1

Gr{oetje,eeting}s,

						Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
							    -- Linus Torvalds

[PATCH/RFC 1/2] ALSA: fireface: Fix integer overflow in transmit_midi_msg()

[Geert Uytterhoeven]

As snd_ff.rx_bytes[] is unsigned int, and NSEC_PER_SEC is 1000000000L,
the second multiplication in

    ff->rx_bytes[port] * 8 * NSEC_PER_SEC / 31250

always overflows on 32-bit platforms, truncating the result.  Fix this
by precalculating "NSEC_PER_SEC / 31250", which is an integer constant.

Note that this assumes ff->rx_bytes[port] <= 16777.

Fixes: 19174295788de77d ("ALSA: fireface: add transaction support")
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
---
Compile-tested only.

I don't know the maximum transfer length of MIDI, but given it's an old
standard, I guess it's rather small.  If it is larger than 16777, the
constant "8" should be replaced by "8ULL", to force 64-bit arithmetic.
---
 sound/firewire/fireface/ff-transaction.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/sound/firewire/fireface/ff-transaction.c b/sound/firewire/fireface/ff-transaction.c
index 7f82762ccc8c80ba..ee7122c461d46f44 100644
--- a/sound/firewire/fireface/ff-transaction.c
+++ b/sound/firewire/fireface/ff-transaction.c
@@ -88,7 +88,7 @@ static void transmit_midi_msg(struct snd_ff *ff, unsigned int port)
 
 	/* Set interval to next transaction. */
 	ff->next_ktime[port] = ktime_add_ns(ktime_get(),
-				ff->rx_bytes[port] * 8 * NSEC_PER_SEC / 31250);
+			ff->rx_bytes[port] * 8 * (NSEC_PER_SEC / 31250));
 
 	if (quad_count == 1)
 		tcode = TCODE_WRITE_QUADLET_REQUEST;
-- 
2.25.1

[PATCH/RFC 2/2] ALSA: firewire-tascam: Fix integer overflow in midi_port_work()

[Geert Uytterhoeven]

As snd_fw_async_midi_port.consume_bytes is unsigned int, and
NSEC_PER_SEC is 1000000000L, the second multiplication in

    port->consume_bytes * 8 * NSEC_PER_SEC / 31250

always overflows on 32-bit platforms, truncating the result.  Fix this
by precalculating "NSEC_PER_SEC / 31250", which is an integer constant.

Note that this assumes port->consume_bytes <= 16777.

Fixes: 531f471834227d03 ("ALSA: firewire-lib/firewire-tascam: localize async midi port")
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
---
Compile-tested only.

I don't know the maximum transfer length of MIDI, but given it's an old
standard, I guess it's rather small.  If it is larger than 16777, the
constant "8" should be replaced by "8ULL", to force 64-bit arithmetic.
---
 sound/firewire/tascam/tascam-transaction.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/sound/firewire/tascam/tascam-transaction.c b/sound/firewire/tascam/tascam-transaction.c
index 90288b4b46379526..a073cece4a7d5e3a 100644
--- a/sound/firewire/tascam/tascam-transaction.c
+++ b/sound/firewire/tascam/tascam-transaction.c
@@ -209,7 +209,7 @@ static void midi_port_work(struct work_struct *work)
 
 	/* Set interval to next transaction. */
 	port->next_ktime = ktime_add_ns(ktime_get(),
-				port->consume_bytes * 8 * NSEC_PER_SEC / 31250);
+			port->consume_bytes * 8 * (NSEC_PER_SEC / 31250));
 
 	/* Start this transaction. */
 	port->idling = false;
-- 
2.25.1

Re: [PATCH/RFC 2/2] ALSA: firewire-tascam: Fix integer overflow in midi_port_work()

[Takashi Sakamoto]

Hi,

On Mon, Jan 11, 2021 at 02:02:51PM +0100, Geert Uytterhoeven wrote:
> As snd_fw_async_midi_port.consume_bytes is unsigned int, and
> NSEC_PER_SEC is 1000000000L, the second multiplication in
> 
>     port->consume_bytes * 8 * NSEC_PER_SEC / 31250
> 
> always overflows on 32-bit platforms, truncating the result.  Fix this
> by precalculating "NSEC_PER_SEC / 31250", which is an integer constant.
> 
> Note that this assumes port->consume_bytes <= 16777.
> 
> Fixes: 531f471834227d03 ("ALSA: firewire-lib/firewire-tascam: localize async midi port")
> Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
> ---
> Compile-tested only.
> 
> I don't know the maximum transfer length of MIDI, but given it's an old
> standard, I guess it's rather small.  If it is larger than 16777, the
> constant "8" should be replaced by "8ULL", to force 64-bit arithmetic.
> ---
>  sound/firewire/tascam/tascam-transaction.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
 
Indeed. The calculation brings integer overflow of 32 bit storage. Thanks
for your care and proposal of the patch. I agree with the intension of
patch, however I have a nitpicking that the consume_bytes member is
defined as 'int', not 'unsigned int' in your commit comment.

The member has value returned from the call of 'fill_message()'[1] for the
length of byte messages in buffer to process, or for error code. The
error code is checked immediately. The range of value is equal to
or less than 3 when reaching the calculation, thus it should be less than
16777.

Regardless of the type of 'int' or 'unsigned int', this patch can fix
the issued problem. Feel free to add my tag when you post second version
with comment fix.

Reviewed-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>

> diff --git a/sound/firewire/tascam/tascam-transaction.c b/sound/firewire/tascam/tascam-transaction.c
> index 90288b4b46379526..a073cece4a7d5e3a 100644
> --- a/sound/firewire/tascam/tascam-transaction.c
> +++ b/sound/firewire/tascam/tascam-transaction.c
> @@ -209,7 +209,7 @@ static void midi_port_work(struct work_struct *work)
>  
>  	/* Set interval to next transaction. */
>  	port->next_ktime = ktime_add_ns(ktime_get(),
> -				port->consume_bytes * 8 * NSEC_PER_SEC / 31250);
> +			port->consume_bytes * 8 * (NSEC_PER_SEC / 31250));
>  
>  	/* Start this transaction. */
>  	port->idling = false;
> -- 
> 2.25.1

[1] https://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound.git/tree/sound/firewire/tascam/tascam-transaction.c#n197

Thanks

Takashi Sakamoto

Re: [PATCH/RFC 1/2] ALSA: fireface: Fix integer overflow in transmit_midi_msg()

[Takashi Sakamoto]

Hi,

On Mon, Jan 11, 2021 at 02:02:50PM +0100, Geert Uytterhoeven wrote:
> As snd_ff.rx_bytes[] is unsigned int, and NSEC_PER_SEC is 1000000000L,
> the second multiplication in
> 
>     ff->rx_bytes[port] * 8 * NSEC_PER_SEC / 31250
> 
> always overflows on 32-bit platforms, truncating the result.  Fix this
> by precalculating "NSEC_PER_SEC / 31250", which is an integer constant.
> 
> Note that this assumes ff->rx_bytes[port] <= 16777.
> 
> Fixes: 19174295788de77d ("ALSA: fireface: add transaction support")
> Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
> ---
> Compile-tested only.
> 
> I don't know the maximum transfer length of MIDI, but given it's an old
> standard, I guess it's rather small.  If it is larger than 16777, the
> constant "8" should be replaced by "8ULL", to force 64-bit arithmetic.
> ---
>  sound/firewire/fireface/ff-transaction.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)

The rx_bytes member has value for the length of byte messages to
process. The range of value differs depending on Fireface protocol
version. For former protocol, the value is equals to or less than
SND_FF_MAXIMIM_MIDI_QUADS (= 9). For latter protocol, the value is
equals to or less than 3. Anyway, the value should not be larger
than 16777 and the calculation can be done without ULL suffix.

Reviewd-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>

> diff --git a/sound/firewire/fireface/ff-transaction.c b/sound/firewire/fireface/ff-transaction.c
> index 7f82762ccc8c80ba..ee7122c461d46f44 100644
> --- a/sound/firewire/fireface/ff-transaction.c
> +++ b/sound/firewire/fireface/ff-transaction.c
> @@ -88,7 +88,7 @@ static void transmit_midi_msg(struct snd_ff *ff, unsigned int port)
>  
>  	/* Set interval to next transaction. */
>  	ff->next_ktime[port] = ktime_add_ns(ktime_get(),
> -				ff->rx_bytes[port] * 8 * NSEC_PER_SEC / 31250);
> +			ff->rx_bytes[port] * 8 * (NSEC_PER_SEC / 31250));
>  
>  	if (quad_count == 1)
>  		tcode = TCODE_WRITE_QUADLET_REQUEST;
> -- 
> 2.25.1


Thanks

Takashi Sakamoto

Re: [PATCH/RFC 2/2] ALSA: firewire-tascam: Fix integer overflow in midi_port_work()

[Geert Uytterhoeven]

Hi Sakamoto-san,

On Tue, Jan 12, 2021 at 2:43 PM Takashi Sakamoto
<o-takashi@sakamocchi.jp> wrote:
> On Mon, Jan 11, 2021 at 02:02:51PM +0100, Geert Uytterhoeven wrote:
> > As snd_fw_async_midi_port.consume_bytes is unsigned int, and
> > NSEC_PER_SEC is 1000000000L, the second multiplication in
> >
> >     port->consume_bytes * 8 * NSEC_PER_SEC / 31250
> >
> > always overflows on 32-bit platforms, truncating the result.  Fix this
> > by precalculating "NSEC_PER_SEC / 31250", which is an integer constant.
> >
> > Note that this assumes port->consume_bytes <= 16777.
> >
> > Fixes: 531f471834227d03 ("ALSA: firewire-lib/firewire-tascam: localize async midi port")
> > Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
> > ---
> > Compile-tested only.
> >
> > I don't know the maximum transfer length of MIDI, but given it's an old
> > standard, I guess it's rather small.  If it is larger than 16777, the
> > constant "8" should be replaced by "8ULL", to force 64-bit arithmetic.
> > ---
> >  sound/firewire/tascam/tascam-transaction.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
>
> Indeed. The calculation brings integer overflow of 32 bit storage. Thanks
> for your care and proposal of the patch. I agree with the intension of
> patch, however I have a nitpicking that the consume_bytes member is
> defined as 'int', not 'unsigned int' in your commit comment.

Thanks, you're right.
Note that port->consume_bytes being signed halves the limit to
8388 bytes, which is of course still met.

> The member has value returned from the call of 'fill_message()'[1] for the
> length of byte messages in buffer to process, or for error code. The
> error code is checked immediately. The range of value is equal to
> or less than 3 when reaching the calculation, thus it should be less than
> 16777.
>
> Regardless of the type of 'int' or 'unsigned int', this patch can fix
> the issued problem. Feel free to add my tag when you post second version
> with comment fix.
>
> Reviewed-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>

Thanks!

Gr{oetje,eeting}s,

                        Geert

-- 
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds

Re: [PATCH/RFC 1/2] ALSA: fireface: Fix integer overflow in transmit_midi_msg()

[Takashi Iwai]

On Mon, 11 Jan 2021 14:02:50 +0100,
Geert Uytterhoeven wrote:
> 
> As snd_ff.rx_bytes[] is unsigned int, and NSEC_PER_SEC is 1000000000L,
> the second multiplication in
> 
>     ff->rx_bytes[port] * 8 * NSEC_PER_SEC / 31250
> 
> always overflows on 32-bit platforms, truncating the result.  Fix this
> by precalculating "NSEC_PER_SEC / 31250", which is an integer constant.
> 
> Note that this assumes ff->rx_bytes[port] <= 16777.
> 
> Fixes: 19174295788de77d ("ALSA: fireface: add transaction support")
> Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
> ---
> Compile-tested only.
> 
> I don't know the maximum transfer length of MIDI, but given it's an old
> standard, I guess it's rather small.  If it is larger than 16777, the
> constant "8" should be replaced by "8ULL", to force 64-bit arithmetic.

Applied now.  Thanks.


Takashi

Re: [PATCH/RFC 2/2] ALSA: firewire-tascam: Fix integer overflow in midi_port_work()

[Takashi Iwai]

On Mon, 11 Jan 2021 14:02:51 +0100,
Geert Uytterhoeven wrote:
> 
> As snd_fw_async_midi_port.consume_bytes is unsigned int, and
> NSEC_PER_SEC is 1000000000L, the second multiplication in
> 
>     port->consume_bytes * 8 * NSEC_PER_SEC / 31250
> 
> always overflows on 32-bit platforms, truncating the result.  Fix this
> by precalculating "NSEC_PER_SEC / 31250", which is an integer constant.
> 
> Note that this assumes port->consume_bytes <= 16777.
> 
> Fixes: 531f471834227d03 ("ALSA: firewire-lib/firewire-tascam: localize async midi port")
> Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
> ---
> Compile-tested only.
> 
> I don't know the maximum transfer length of MIDI, but given it's an old
> standard, I guess it's rather small.  If it is larger than 16777, the
> constant "8" should be replaced by "8ULL", to force 64-bit arithmetic.

Applied now.  Thanks.


Takashi