2016-10-31 15:13:52

by Maarten ter Huurne

Subject: [PATCH] dma-debug: Avoid NULL dereference when checking sync

check_sync() calls bucket_find_contain(), which in turn calls
dma_get_max_seg_size(), which dereferences the device pointer.

Signed-off-by: Maarten ter Huurne <[email protected]>
---
lib/dma-debug.c | 5 +++++
1 file changed, 5 insertions(+)

diff --git a/lib/dma-debug.c b/lib/dma-debug.c
index 8971370..84c6e88 100644
--- a/lib/dma-debug.c
+++ b/lib/dma-debug.c
@@ -1223,6 +1223,11 @@ static void check_sync(struct device *dev,
struct hash_bucket *bucket;
unsigned long flags;

+ if (!ref->dev) {
+ err_printk(dev, NULL, "DMA-API: device driver passes NULL for device to DMA sync function; cannot check usage\n");
+ return;
+ }
+
bucket = get_hash_bucket(ref, &flags);

entry = bucket_find_contain(&bucket, ref, &flags);
--
2.6.6

2016-11-08 00:39:02

by Andrew Morton

Subject: Re: [PATCH] dma-debug: Avoid NULL dereference when checking sync

On Mon, 31 Oct 2016 16:11:41 +0100 Maarten ter Huurne <[email protected]> wrote:

> check_sync() calls bucket_find_contain(), which in turn calls
> dma_get_max_seg_size(), which dereferences the device pointer.
>
> ...
>
> --- a/lib/dma-debug.c
> +++ b/lib/dma-debug.c
> @@ -1223,6 +1223,11 @@ static void check_sync(struct device *dev,
> struct hash_bucket *bucket;
> unsigned long flags;
>
> + if (!ref->dev) {
> + err_printk(dev, NULL, "DMA-API: device driver passes NULL for device to DMA sync function; cannot check usage\n");
> + return;
> + }
> +
> bucket = get_hash_bucket(ref, &flags);
>
> entry = bucket_find_contain(&bucket, ref, &flags);

Well.

a) who is calling check_sync() without having filled in ref.dev?
How does this occur? Is this a bug in the caller?

b) why does check_sync() have a `dev' argument anyway? Is there any
situation in which that will differ from ref.dev?