[PATCH] dmaengine: sun6i: fix use-after-free in descriptor error path

[PATCH] dmaengine: sun6i: fix use-after-free in descriptor error path

[Chaithanya Lagisetty]

The err_lli_free cleanup loops in sun6i_dma_prep_slave_sg() and
sun6i_dma_prep_dma_cyclic() free each LLI node with dma_pool_free()
and then dereference the freed node to obtain v_lli_next and
p_lli_next for the next loop iteration.

Save the next pointers before freeing, matching the pattern already used
in sun6i_dma_free_desc().

Fixes: 9aa48806edb8 ("dmaengine: sun6i: Do not use virt_to_phys")
Signed-off-by: Chaithanya Lagisetty <nagachaithanya9911@gmail.com>
---
 drivers/dma/sun6i-dma.c | 18 ++++++++++++------
 1 file changed, 12 insertions(+), 6 deletions(-)

diff --git a/drivers/dma/sun6i-dma.c b/drivers/dma/sun6i-dma.c
index f47a326dd7ff..2439fa6564f2 100644
--- a/drivers/dma/sun6i-dma.c
+++ b/drivers/dma/sun6i-dma.c
@@ -720,10 +720,10 @@ static struct dma_async_tx_descriptor *sun6i_dma_prep_slave_sg(
 	struct sun6i_dma_dev *sdev = to_sun6i_dma_dev(chan->device);
 	struct sun6i_vchan *vchan = to_sun6i_vchan(chan);
 	struct dma_slave_config *sconfig = &vchan->cfg;
-	struct sun6i_dma_lli *v_lli, *prev = NULL;
+	struct sun6i_dma_lli *v_lli, *v_next, *prev = NULL;
 	struct sun6i_desc *txd;
 	struct scatterlist *sg;
-	dma_addr_t p_lli;
+	dma_addr_t p_lli, p_next;
 	u32 lli_cfg;
 	int i, ret;
 
@@ -789,8 +789,11 @@ static struct dma_async_tx_descriptor *sun6i_dma_prep_slave_sg(
 
 err_lli_free:
 	for (p_lli = txd->p_lli, v_lli = txd->v_lli; v_lli;
-	     p_lli = v_lli->p_lli_next, v_lli = v_lli->v_lli_next)
+	     p_lli = p_next, v_lli = v_next) {
+		v_next = v_lli->v_lli_next;
+		p_next = v_lli->p_lli_next;
 		dma_pool_free(sdev->pool, v_lli, p_lli);
+	}
 	kfree(txd);
 	return NULL;
 }
@@ -806,9 +809,9 @@ static struct dma_async_tx_descriptor *sun6i_dma_prep_dma_cyclic(
 	struct sun6i_dma_dev *sdev = to_sun6i_dma_dev(chan->device);
 	struct sun6i_vchan *vchan = to_sun6i_vchan(chan);
 	struct dma_slave_config *sconfig = &vchan->cfg;
-	struct sun6i_dma_lli *v_lli, *prev = NULL;
+	struct sun6i_dma_lli *v_lli, *v_next, *prev = NULL;
 	struct sun6i_desc *txd;
-	dma_addr_t p_lli;
+	dma_addr_t p_lli, p_next;
 	u32 lli_cfg;
 	unsigned int i, periods = buf_len / period_len;
 	int ret;
@@ -870,8 +873,11 @@ static struct dma_async_tx_descriptor *sun6i_dma_prep_dma_cyclic(
 
 err_lli_free:
 	for (p_lli = txd->p_lli, v_lli = txd->v_lli; v_lli;
-	     p_lli = v_lli->p_lli_next, v_lli = v_lli->v_lli_next)
+	     p_lli = p_next, v_lli = v_next) {
+		v_next = v_lli->v_lli_next;
+		p_next = v_lli->p_lli_next;
 		dma_pool_free(sdev->pool, v_lli, p_lli);
+	}
 	kfree(txd);
 	return NULL;
 }
-- 
2.43.0

Re: [PATCH] dmaengine: sun6i: fix use-after-free in descriptor error path

[sashiko-bot]

Thank you for your contribution! Sashiko AI review found 3 potential issue(s) to consider:

Pre-existing issues:
- [High] Missing validation for `period_len` and `buf_len` in `sun6i_dma_prep_dma_cyclic` leads to a division by zero or a NULL pointer dereference.
- [High] Missing validation for `sg_len == 0` in `sun6i_dma_prep_slave_sg` leads to hardware being programmed with physical address 0x0.
- [High] Integer truncation in the interrupt handler loop bound causes interrupts from channels >= 8 to be permanently ignored on SoCs with non-multiple-of-8 channels (e.g., sun8i-h3).
--

--- Patch [1]: [PATCH] dmaengine: sun6i: fix use-after-free in descriptor error path ---
commit 815343fddc5f554759b857f782dbda4a7c8b1a7a
Author: Chaithanya Lagisetty <nagachaithanya9911@gmail.com>

dmaengine: sun6i: fix use-after-free in descriptor error path

The err_lli_free cleanup loops in sun6i_dma_prep_slave_sg() and
sun6i_dma_prep_dma_cyclic() free each LLI node with dma_pool_free()
and then dereference the freed node to obtain v_lli_next and
p_lli_next for the next loop iteration.

> diff --git a/drivers/dma/sun6i-dma.c b/drivers/dma/sun6i-dma.c
> index a9a254dbf8cb9..2c163ab22af5c 100644
> --- a/drivers/dma/sun6i-dma.c
> +++ b/drivers/dma/sun6i-dma.c
> @@ -720,10 +720,10 @@ static struct dma_async_tx_descriptor *sun6i_dma_prep_slave_sg(
>  	struct sun6i_dma_dev *sdev = to_sun6i_dma_dev(chan->device);
>  	struct sun6i_vchan *vchan = to_sun6i_vchan(chan);
>  	struct dma_slave_config *sconfig = &vchan->cfg;
> -	struct sun6i_dma_lli *v_lli, *prev = NULL;
> +	struct sun6i_dma_lli *v_lli, *v_next, *prev = NULL;
>  	struct sun6i_desc *txd;
>  	struct scatterlist *sg;
> -	dma_addr_t p_lli;
> +	dma_addr_t p_lli, p_next;
>  	u32 lli_cfg;
>  	int i, ret;
>  

[Severity: High]
This is a pre-existing issue, but I noticed there is no validation for
sg_len == 0.

If a caller provides an empty scatterlist, the txd is allocated but the
for_each_sg() loop is skipped:

drivers/dma/sun6i-dma.c:sun6i_dma_prep_slave_sg() {
    ...
    for_each_sg(sgl, sg, sg_len, i) {
        ...
    }
    ...
}

This leaves txd->p_lli as 0 and txd->v_lli as NULL. When this descriptor is
executed via sun6i_dma_start_desc(), the hardware is programmed with physical
address 0x0.

Could this lead to the hardware accessing physical address 0x0?

[ ... ]

> @@ -806,9 +809,9 @@ static struct dma_async_tx_descriptor *sun6i_dma_prep_dma_cyclic(
>  	struct sun6i_dma_dev *sdev = to_sun6i_dma_dev(chan->device);
>  	struct sun6i_vchan *vchan = to_sun6i_vchan(chan);
>  	struct dma_slave_config *sconfig = &vchan->cfg;
> -	struct sun6i_dma_lli *v_lli, *prev = NULL;
> +	struct sun6i_dma_lli *v_lli, *v_next, *prev = NULL;
>  	struct sun6i_desc *txd;
> -	dma_addr_t p_lli;
> +	dma_addr_t p_lli, p_next;
>  	u32 lli_cfg;
>  	unsigned int i, periods = buf_len / period_len;

[Severity: High]
This is a pre-existing issue, but is it possible for period_len to be 0 here,
leading to a division by zero?

Additionally, if buf_len < period_len, periods evaluates to 0, which skips
the allocation loop. This leaves prev as NULL, and causes a NULL pointer
dereference later in the function:

drivers/dma/sun6i-dma.c:sun6i_dma_prep_dma_cyclic() {
    ...
    prev->p_lli_next = txd->p_lli;      /* cyclic list */
    ...
}

Should we validate period_len and buf_len before performing this division?

[ ... ]

While reviewing this code, I also noticed another pre-existing issue in the
interrupt handler.

[Severity: High]
This is a pre-existing issue, but I noticed a potential truncation in the
channel group calculation for the interrupt handler.

drivers/dma/sun6i-dma.c:sun6i_dma_interrupt() {
    ...
    for (i = 0; i < sdev->num_pchans / DMA_IRQ_CHAN_NR; i++) {
        ...
    }
    ...
}

For SoCs with non-multiple-of-8 channel counts (like sun8i-h3 with 12
channels), 12 / 8 evaluates to 1. This causes the handler to only process
channels 0-7 and completely skip channels 8-11.

Could this result in unhandled interrupts and system hard lockups if a DMA
transfer completes on channels 8-11?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260810102044.1725649-1-nagachaithanya9911@gmail.com?part=1