diff mbox series

[v2] rtw88: Fix out-of-bounds write

Message ID 20210716155311.5570-1-len.baker@gmx.com
State New
Headers show
Series [v2] rtw88: Fix out-of-bounds write | expand

Commit Message

Len Baker July 16, 2021, 3:53 p.m. UTC
In the rtw_pci_init_rx_ring function the "if (len > TRX_BD_IDX_MASK)"
statement guarantees that len is less than or equal to GENMASK(11, 0) or
in other words that len is less than or equal to 4095. However the
rx_ring->buf has a size of RTK_MAX_RX_DESC_NUM (defined as 512). This
way it is possible an out-of-bounds write in the for statement due to
the i variable can exceed the rx_ring->buff size.

However, this overflow never happens due to the rtw_pci_init_rx_ring is
only ever called with a fixed constant of RTK_MAX_RX_DESC_NUM. But it is
better to be defensive in this case and add a new check to avoid
overflows if this function is called in a future with a value greater
than 512.

Cc: stable@vger.kernel.org
Addresses-Coverity-ID: 1461515 ("Out-of-bounds write")
Fixes: e3037485c68ec ("rtw88: new Realtek 802.11ac driver")
Signed-off-by: Len Baker <len.baker@gmx.com>
---
Changelog v1 -> v2
- Remove the macro ARRAY_SIZE from the for loop (Pkshih, Brian Norris).
- Add a new check for the len variable (Pkshih, Brian Norris).

 drivers/net/wireless/realtek/rtw88/pci.c | 5 +++++
 1 file changed, 5 insertions(+)

--
2.25.1

Comments

Greg KH July 16, 2021, 5:20 p.m. UTC | #1
On Fri, Jul 16, 2021 at 05:53:11PM +0200, Len Baker wrote:
> In the rtw_pci_init_rx_ring function the "if (len > TRX_BD_IDX_MASK)"
> statement guarantees that len is less than or equal to GENMASK(11, 0) or
> in other words that len is less than or equal to 4095. However the
> rx_ring->buf has a size of RTK_MAX_RX_DESC_NUM (defined as 512). This
> way it is possible an out-of-bounds write in the for statement due to
> the i variable can exceed the rx_ring->buff size.
> 
> However, this overflow never happens due to the rtw_pci_init_rx_ring is
> only ever called with a fixed constant of RTK_MAX_RX_DESC_NUM. But it is
> better to be defensive in this case and add a new check to avoid
> overflows if this function is called in a future with a value greater
> than 512.

If this can never happen, then no, this is not needed.  Why would you
check twice for the same thing?

thanks,

greg k-h
Len Baker July 17, 2021, 1:33 p.m. UTC | #2
On Fri, Jul 16, 2021 at 07:20:48PM +0200, Greg KH wrote:
> On Fri, Jul 16, 2021 at 05:53:11PM +0200, Len Baker wrote:
> > In the rtw_pci_init_rx_ring function the "if (len > TRX_BD_IDX_MASK)"
> > statement guarantees that len is less than or equal to GENMASK(11, 0) or
> > in other words that len is less than or equal to 4095. However the
> > rx_ring->buf has a size of RTK_MAX_RX_DESC_NUM (defined as 512). This
> > way it is possible an out-of-bounds write in the for statement due to
> > the i variable can exceed the rx_ring->buff size.
> >
> > However, this overflow never happens due to the rtw_pci_init_rx_ring is
> > only ever called with a fixed constant of RTK_MAX_RX_DESC_NUM. But it is
> > better to be defensive in this case and add a new check to avoid
> > overflows if this function is called in a future with a value greater
> > than 512.
>
> If this can never happen, then no, this is not needed.

Then, if this can never happen, the current check would not be necessary
either.

> Why would you check twice for the same thing?

Ok, it makes no sense to double check the "len" variable twice. So, I
propose to modify the current check as follows:

diff --git a/drivers/net/wireless/realtek/rtw88/pci.c b/drivers/net/wireless/realtek/rtw88/pci.c
index e7d17ab8f113..0fd140523868 100644
--- a/drivers/net/wireless/realtek/rtw88/pci.c
+++ b/drivers/net/wireless/realtek/rtw88/pci.c
@@ -268,8 +268,8 @@ static int rtw_pci_init_rx_ring(struct rtw_dev *rtwdev,
        int i, allocated;
        int ret = 0;

-       if (len > TRX_BD_IDX_MASK) {
-               rtw_err(rtwdev, "len %d exceeds maximum RX entries\n", len);
+       if (len > ARRAY_SIZE(rx_ring->buf)) {
+               rtw_err(rtwdev, "len %d exceeds maximum RX ring buffer\n", len);
                return -EINVAL;
        }

This way the overflow can never happen with the current call to
rtw_pci_init_rx_ring function or with a future call with a "len" parameter
greater than 512. What do you think?

If there are no objections I will send a v3 for review.

Another question: If this can never happen should I include the "Fixes" tag,
"Addresses-Coverity-ID" tag and Cc to stable?

Thanks,
Len

>
> thanks,
>
> greg k-h
Greg KH July 17, 2021, 5:33 p.m. UTC | #3
On Sat, Jul 17, 2021 at 03:33:43PM +0200, Len Baker wrote:
> On Fri, Jul 16, 2021 at 07:20:48PM +0200, Greg KH wrote:
> > On Fri, Jul 16, 2021 at 05:53:11PM +0200, Len Baker wrote:
> > > In the rtw_pci_init_rx_ring function the "if (len > TRX_BD_IDX_MASK)"
> > > statement guarantees that len is less than or equal to GENMASK(11, 0) or
> > > in other words that len is less than or equal to 4095. However the
> > > rx_ring->buf has a size of RTK_MAX_RX_DESC_NUM (defined as 512). This
> > > way it is possible an out-of-bounds write in the for statement due to
> > > the i variable can exceed the rx_ring->buff size.
> > >
> > > However, this overflow never happens due to the rtw_pci_init_rx_ring is
> > > only ever called with a fixed constant of RTK_MAX_RX_DESC_NUM. But it is
> > > better to be defensive in this case and add a new check to avoid
> > > overflows if this function is called in a future with a value greater
> > > than 512.
> >
> > If this can never happen, then no, this is not needed.
> 
> Then, if this can never happen, the current check would not be necessary
> either.
> 
> > Why would you check twice for the same thing?
> 
> Ok, it makes no sense to double check the "len" variable twice. So, I
> propose to modify the current check as follows:
> 
> diff --git a/drivers/net/wireless/realtek/rtw88/pci.c b/drivers/net/wireless/realtek/rtw88/pci.c
> index e7d17ab8f113..0fd140523868 100644
> --- a/drivers/net/wireless/realtek/rtw88/pci.c
> +++ b/drivers/net/wireless/realtek/rtw88/pci.c
> @@ -268,8 +268,8 @@ static int rtw_pci_init_rx_ring(struct rtw_dev *rtwdev,
>         int i, allocated;
>         int ret = 0;
> 
> -       if (len > TRX_BD_IDX_MASK) {
> -               rtw_err(rtwdev, "len %d exceeds maximum RX entries\n", len);
> +       if (len > ARRAY_SIZE(rx_ring->buf)) {
> +               rtw_err(rtwdev, "len %d exceeds maximum RX ring buffer\n", len);
>                 return -EINVAL;
>         }
> 
> This way the overflow can never happen with the current call to
> rtw_pci_init_rx_ring function or with a future call with a "len" parameter
> greater than 512. What do you think?
> 
> If there are no objections I will send a v3 for review.
> 
> Another question: If this can never happen should I include the "Fixes" tag,
> "Addresses-Coverity-ID" tag and Cc to stable?

If it can never happen, why have this check at all?

Looks like a Coverity false positive?

thanks,

greg k-h
Len Baker July 18, 2021, 7:53 a.m. UTC | #4
On Sat, Jul 17, 2021 at 07:33:49PM +0200, Greg KH wrote:
> On Sat, Jul 17, 2021 at 03:33:43PM +0200, Len Baker wrote:

> > Another question: If this can never happen should I include the "Fixes" tag,

> > "Addresses-Coverity-ID" tag and Cc to stable?

>

> If it can never happen, why have this check at all?

>

> Looks like a Coverity false positive?


Ok, then I will remove the check and I will send a patch for review.

>

> thanks,

>

> greg k-h


Regards,
Len
diff mbox series

Patch

diff --git a/drivers/net/wireless/realtek/rtw88/pci.c b/drivers/net/wireless/realtek/rtw88/pci.c
index e7d17ab8f113..53dc90276693 100644
--- a/drivers/net/wireless/realtek/rtw88/pci.c
+++ b/drivers/net/wireless/realtek/rtw88/pci.c
@@ -273,6 +273,11 @@  static int rtw_pci_init_rx_ring(struct rtw_dev *rtwdev,
 		return -EINVAL;
 	}

+	if (len > ARRAY_SIZE(rx_ring->buf)) {
+		rtw_err(rtwdev, "len %d exceeds maximum RX ring buffer\n", len);
+		return -EINVAL;
+	}
+
 	head = dma_alloc_coherent(&pdev->dev, ring_sz, &dma, GFP_KERNEL);
 	if (!head) {
 		rtw_err(rtwdev, "failed to allocate rx ring\n");