Re: [PATCH 2/2] mtd: nand: qpic_common: prevent out of bounds access of BAM arrays

From: Miquel Raynal <miquel.raynal@bootlin.com>
To: Gabor Juhos <j4g8y7@gmail.com>
Cc: Mark Brown <broonie@kernel.org>,
	 Md Sadre Alam <quic_mdalam@quicinc.com>,
	 Varadarajan Narayanan <quic_varada@quicinc.com>,
	 Sricharan Ramabadhran <quic_srichara@quicinc.com>,
	 Richard Weinberger <richard@nod.at>,
	Vignesh Raghavendra <vigneshr@ti.com>,
	 linux-spi@vger.kernel.org, linux-mtd@lists.infradead.org,
	 linux-arm-msm@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH 2/2] mtd: nand: qpic_common: prevent out of bounds access of BAM arrays
Date: Mon, 26 May 2025 11:27:45 +0200	[thread overview]
Message-ID: <87cybv2032.fsf@bootlin.com> (raw)
In-Reply-To: <20250525-qpic-snand-avoid-mem-corruption-v1-2-5fe528def7fb@gmail.com> (Gabor Juhos's message of "Sun, 25 May 2025 19:05:36 +0200")

Hi Gabor,

On 25/05/2025 at 19:05:36 +02, Gabor Juhos <j4g8y7@gmail.com> wrote:

> The common QPIC code does not do any boundary checking when it handles
> the command elements and scatter gater list arrays of a BAM transaction,
> thus it allows to access out of bounds elements in those.
>
> Although it is the responsibility of the given driver to allocate enough
> space for all possible BAM transaction variations, however there can be
> mistakes in the driver code which can lead to hidden memory corruption
> issues which are hard to debug.
>
> This kind of problem has been observed during testing the 'spi-qpic-snand'
> driver. Although the driver has been fixed with a preceding patch, but it
> still makes sense to reduce the chance of having such errors again later.
>
> In order to prevent such errors, change the qcom_alloc_bam_transaction()
> function to store the number of elements of the arrays in the
> 'bam_transaction' strucutre during allocation. Also, add sanity checks to
> the qcom_prep_bam_dma_desc_{cmd,data}() functions to avoid using out of
> bounds indices for the arrays.
>
> Tested with the 'spi-qpic-snand' driver only.
>
> Signed-off-by: Gabor Juhos <j4g8y7@gmail.com>
> ---
> Preferably, this should go in via the SPI tree along with the previous
> patch. It is not a strict requirement though, in the case it gets
> included separately through the mtd tree it reveals the bug fixed in
> the first patch.

Sorry, didn't see that in the first place. Fine by me.

> ---
>  drivers/mtd/nand/qpic_common.c       | 28 ++++++++++++++++++++++++----
>  include/linux/mtd/nand-qpic-common.h |  8 ++++++++
>  2 files changed, 32 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/mtd/nand/qpic_common.c b/drivers/mtd/nand/qpic_common.c
> index e0ed25b5afea9b289b767cd3d9c2d7572ed52008..fb1f81e4bdacaa3e81660a20e164926c64633513 100644
> --- a/drivers/mtd/nand/qpic_common.c
> +++ b/drivers/mtd/nand/qpic_common.c
> @@ -15,6 +15,13 @@
>  #include <linux/slab.h>
>  #include <linux/mtd/nand-qpic-common.h>
>  
> +static inline int qcom_err_bam_array_full(struct qcom_nand_controller *nandc,
> +					  const char *name)
> +{
> +	dev_err(nandc->dev, "BAM %s array is full\n", name);
> +	return -EINVAL;
> +}

This is rather uncommon, I don't know if it's very relevant to do
that. Please drop this static inline function.

Thanks,
Miquèl