2020-05-30 02:46:53

by Jia-Ju Bai

Subject: [PATCH] media: venus: fix possible buffer overlow casued bad DMA value in venus_sfr_print()

The value hdev->sfr.kva is stored in DMA memory, and it is assigned to
sfr, so sfr->buf_size can be modified at anytime by malicious hardware.
In this case, a buffer overflow may happen when the code
"sfr->data[sfr->buf_size - 1]" is executed.

To fix this possible bug, sfr->buf_size is assigned to a local variable,
and then this variable is checked before being used.

Signed-off-by: Jia-Ju Bai <[email protected]>
---
drivers/media/platform/qcom/venus/hfi_venus.c | 9 +++++++--
1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/drivers/media/platform/qcom/venus/hfi_venus.c b/drivers/media/platform/qcom/venus/hfi_venus.c
index 0d8855014ab3..4251a9e47a1b 100644
--- a/drivers/media/platform/qcom/venus/hfi_venus.c
+++ b/drivers/media/platform/qcom/venus/hfi_venus.c
@@ -960,18 +960,23 @@ static void venus_sfr_print(struct venus_hfi_device *hdev)
{
struct device *dev = hdev->core->dev;
struct hfi_sfr *sfr = hdev->sfr.kva;
+ u32 buf_size;
void *p;

if (!sfr)
return;

- p = memchr(sfr->data, '\0', sfr->buf_size);
+ buf_size = sfr->buf_size;
+ if (buf_size > 1)
+ return;
+
+ p = memchr(sfr->data, '\0', buf_size);
/*
* SFR isn't guaranteed to be NULL terminated since SYS_ERROR indicates
* that Venus is in the process of crashing.
*/
if (!p)
- sfr->data[sfr->buf_size - 1] = '\0';
+ sfr->data[buf_size - 1] = '\0';

dev_err_ratelimited(dev, "SFR message from FW: %s\n", sfr->data);
}
--
2.17.1

2020-11-27 11:18:14

by Mauro Carvalho Chehab

Subject: Re: [PATCH] media: venus: fix possible buffer overlow casued bad DMA value in venus_sfr_print()

Em Sat, 30 May 2020 10:41:17 +0800
Jia-Ju Bai <[email protected]> escreveu:

> The value hdev->sfr.kva is stored in DMA memory, and it is assigned to
> sfr, so sfr->buf_size can be modified at anytime by malicious hardware.
> In this case, a buffer overflow may happen when the code
> "sfr->data[sfr->buf_size - 1]" is executed.
>
> To fix this possible bug, sfr->buf_size is assigned to a local variable,
> and then this variable is checked before being used.
>
> Signed-off-by: Jia-Ju Bai <[email protected]>
> ---
> drivers/media/platform/qcom/venus/hfi_venus.c | 9 +++++++--
> 1 file changed, 7 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/media/platform/qcom/venus/hfi_venus.c b/drivers/media/platform/qcom/venus/hfi_venus.c
> index 0d8855014ab3..4251a9e47a1b 100644
> --- a/drivers/media/platform/qcom/venus/hfi_venus.c
> +++ b/drivers/media/platform/qcom/venus/hfi_venus.c
> @@ -960,18 +960,23 @@ static void venus_sfr_print(struct venus_hfi_device *hdev)
> {
> struct device *dev = hdev->core->dev;
> struct hfi_sfr *sfr = hdev->sfr.kva;
> + u32 buf_size;
> void *p;
>
> if (!sfr)
> return;
>
> - p = memchr(sfr->data, '\0', sfr->buf_size);
> + buf_size = sfr->buf_size;
> + if (buf_size > 1)

That seems plain wrong to me... I suspect you wanted to do,
instead:

if (buf_size < 1)

or even:
if (buf_size < 1 || buf_size >= maximum_size_of_data)


> + return;
> +
> + p = memchr(sfr->data, '\0', buf_size);
> /*
> * SFR isn't guaranteed to be NULL terminated since SYS_ERROR indicates
> * that Venus is in the process of crashing.
> */
> if (!p)
> - sfr->data[sfr->buf_size - 1] = '\0';
> + sfr->data[buf_size - 1] = '\0';

Well, a malicious hardware with DMA access could simply write 0 to
some random address, without needing to rely on the value
of sfr->buf_size. I can't see how a change like that would prevent
that.

A check like that only makes sense if the driver can ever
call this function with an invalid value for sfr->buf_size.


>
> dev_err_ratelimited(dev, "SFR message from FW: %s\n", sfr->data);
> }



Thanks,
Mauro