2023-10-03 13:13:55

by Loic Poulain

Subject: [PATCH] nvmem: core: Fix possible buffer overflow on nvmem cell write

Nothing prevents a nvmem consumer to try writing excessive data to a
given nvmem cell (except when bit_offset is 0). The allocated buffer
of size 'cell->bytes' in nvmem_cell_prepare_write_buffer may not be
large enough to host the copied 'len' bytes.

Signed-off-by: Loic Poulain <[email protected]>
---
drivers/nvmem/core.c | 3 +++
1 file changed, 3 insertions(+)

diff --git a/drivers/nvmem/core.c b/drivers/nvmem/core.c
index eaf6a3fe8ca6..0b27ab3b3b86 100644
--- a/drivers/nvmem/core.c
+++ b/drivers/nvmem/core.c
@@ -1654,6 +1654,9 @@ static void *nvmem_cell_prepare_write_buffer(struct nvmem_cell_entry *cell,
int i, rc, nbits, bit_offset = cell->bit_offset;
u8 v, *p, *buf, *b, pbyte, pbits;

+ if (len > cell->bytes)
+ return ERR_PTR(-EINVAL);
+
nbits = cell->nbits;
buf = kzalloc(cell->bytes, GFP_KERNEL);
if (!buf)
--
2.34.1

2023-10-07 10:23:00

by Srinivas Kandagatla

Subject: Re: [PATCH] nvmem: core: Fix possible buffer overflow on nvmem cell write

Thanks Loic for the patch,

On 03/10/2023 14:13, Loic Poulain wrote:
> Nothing prevents a nvmem consumer to try writing excessive data to a
> given nvmem cell (except when bit_offset is 0). The allocated buffer
> of size 'cell->bytes' in nvmem_cell_prepare_write_buffer may not be
> large enough to host the copied 'len' bytes.
>
Did you hit this path?

__nvmem_cell_entry_write already has a check for (cell->bit_offset ==
0 && len != cell->bytes))

What is the bit_offset in your case?

Can you provide more details?

thanks,
srini

> Signed-off-by: Loic Poulain <[email protected]>
> ---
> drivers/nvmem/core.c | 3 +++
> 1 file changed, 3 insertions(+)
>
> diff --git a/drivers/nvmem/core.c b/drivers/nvmem/core.c
> index eaf6a3fe8ca6..0b27ab3b3b86 100644
> --- a/drivers/nvmem/core.c
> +++ b/drivers/nvmem/core.c
> @@ -1654,6 +1654,9 @@ static void *nvmem_cell_prepare_write_buffer(struct nvmem_cell_entry *cell,
> int i, rc, nbits, bit_offset = cell->bit_offset;
> u8 v, *p, *buf, *b, pbyte, pbits;
>
> + if (len > cell->bytes)
> + return ERR_PTR(-EINVAL);
> +
> nbits = cell->nbits;
> buf = kzalloc(cell->bytes, GFP_KERNEL);
> if (!buf)

2023-10-16 13:50:33

by Loic Poulain

Subject: Re: [PATCH] nvmem: core: Fix possible buffer overflow on nvmem cell write

Hi Srini,

On Sat, 7 Oct 2023 at 12:22, Srinivas Kandagatla
<[email protected]> wrote:
>
> Thanks Loic for the patch,
>
> On 03/10/2023 14:13, Loic Poulain wrote:
> > Nothing prevents a nvmem consumer to try writing excessive data to a
> > given nvmem cell (except when bit_offset is 0). The allocated buffer
> > of size 'cell->bytes' in nvmem_cell_prepare_write_buffer may not be
> > large enough to host the copied 'len' bytes.
> >
> Did you hit this path?
>
> __nvmem_cell_entry_write already has a check for (cell->bit_offset ==
> 0 && len != cell->bytes))
>
> What is the bit_offset in your case?
>
> Can you provide more details?

I hit the issue while playing with nvmem-reboot-mode driver,
allocating 2-bit of a persistent register at bit-offset 2 for the
reboot mode. nvmem-reboot-mode drivers call nvmem_cell_write() with a
32-bit len value, so we end in nvmem_cell_prepare_write_buffer
allocating a 1-byte (cell->bytes) buffer and copying a 4-byte len
value into it. You can find below the dts example.

```
{
&snvs_lpgpr{
#address-cells = <1>;
#size-cells = <1>;

something@0 {
/* reg[2:0] */
reg = <0x0 0x4>;
bits = <2 2>;
};

reboot_mode: reboot-mode@0 {
/* reg[4:2] */
reg = <0x0 0x4>;
bits = <2 2>;
};
};

reboot-mode {
compatible = "nvmem-reboot-mode";
nvmem-cells = <&reboot_mode>;
nvmem-cell-names = "reboot-mode";
mode-normal = <0>;
mode-fastboot = <1>;
mode-recovery = <2>;
};
};
```