[PATCH 1/1] v4l: mem2mem_testdev: Fix race conditions in driver.

[Tomasz MoÅ]

The mem2mem_testdev allows multiple instances to be opened in parallel.
Source and destination queue data are being shared between all
instances, which can lead to kernel oops due to race conditions (most
likely to happen inside device_run()).

Attached patch fixes mentioned problem by storing queue data per device
context.

Signed-off-by: Tomasz MoÅ <desowin@xxxxxxxxx>
---
 drivers/media/video/mem2mem_testdev.c |   50 +++++++++++++++++----------------
 1 file changed, 26 insertions(+), 24 deletions(-)

diff --git a/drivers/media/video/mem2mem_testdev.c b/drivers/media/video/mem2mem_testdev.c
index 12897e8..ae7ca12 100644
--- a/drivers/media/video/mem2mem_testdev.c
+++ b/drivers/media/video/mem2mem_testdev.c
@@ -110,22 +110,6 @@ enum {
 	V4L2_M2M_DST = 1,
 };
 
-/* Source and destination queue data */
-static struct m2mtest_q_data q_data[2];
-
-static struct m2mtest_q_data *get_q_data(enum v4l2_buf_type type)
-{
-	switch (type) {
-	case V4L2_BUF_TYPE_VIDEO_OUTPUT:
-		return &q_data[V4L2_M2M_SRC];
-	case V4L2_BUF_TYPE_VIDEO_CAPTURE:
-		return &q_data[V4L2_M2M_DST];
-	default:
-		BUG();
-	}
-	return NULL;
-}
-
 #define V4L2_CID_TRANS_TIME_MSEC	V4L2_CID_PRIVATE_BASE
 #define V4L2_CID_TRANS_NUM_BUFS		(V4L2_CID_PRIVATE_BASE + 1)
 
@@ -198,8 +182,26 @@ struct m2mtest_ctx {
 	int			aborting;
 
 	struct v4l2_m2m_ctx	*m2m_ctx;
+
+	/* Source and destination queue data */
+	struct m2mtest_q_data   q_data[2];
 };
 
+static struct m2mtest_q_data *get_q_data(struct m2mtest_ctx *ctx,
+					 enum v4l2_buf_type type)
+{
+	switch (type) {
+	case V4L2_BUF_TYPE_VIDEO_OUTPUT:
+		return &ctx->q_data[V4L2_M2M_SRC];
+	case V4L2_BUF_TYPE_VIDEO_CAPTURE:
+		return &ctx->q_data[V4L2_M2M_DST];
+	default:
+		BUG();
+	}
+	return NULL;
+}
+
+
 static struct v4l2_queryctrl *get_ctrl(int id)
 {
 	int i;
@@ -223,7 +225,7 @@ static int device_process(struct m2mtest_ctx *ctx,
 	int tile_w, bytes_left;
 	int width, height, bytesperline;
 
-	q_data = get_q_data(V4L2_BUF_TYPE_VIDEO_OUTPUT);
+	q_data = get_q_data(ctx, V4L2_BUF_TYPE_VIDEO_OUTPUT);
 
 	width	= q_data->width;
 	height	= q_data->height;
@@ -436,7 +438,7 @@ static int vidioc_g_fmt(struct m2mtest_ctx *ctx, struct v4l2_format *f)
 	if (!vq)
 		return -EINVAL;
 
-	q_data = get_q_data(f->type);
+	q_data = get_q_data(ctx, f->type);
 
 	f->fmt.pix.width	= q_data->width;
 	f->fmt.pix.height	= q_data->height;
@@ -535,7 +537,7 @@ static int vidioc_s_fmt(struct m2mtest_ctx *ctx, struct v4l2_format *f)
 	if (!vq)
 		return -EINVAL;
 
-	q_data = get_q_data(f->type);
+	q_data = get_q_data(ctx, f->type);
 	if (!q_data)
 		return -EINVAL;
 
@@ -747,7 +749,7 @@ static int m2mtest_queue_setup(struct vb2_queue *vq,
 	struct m2mtest_q_data *q_data;
 	unsigned int size, count = *nbuffers;
 
-	q_data = get_q_data(vq->type);
+	q_data = get_q_data(ctx, vq->type);
 
 	size = q_data->width * q_data->height * q_data->fmt->depth >> 3;
 
@@ -775,7 +777,7 @@ static int m2mtest_buf_prepare(struct vb2_buffer *vb)
 
 	dprintk(ctx->dev, "type: %d\n", vb->vb2_queue->type);
 
-	q_data = get_q_data(vb->vb2_queue->type);
+	q_data = get_q_data(ctx, vb->vb2_queue->type);
 
 	if (vb2_plane_size(vb, 0) < q_data->sizeimage) {
 		dprintk(ctx->dev, "%s data will not fit into plane (%lu < %lu)\n",
@@ -860,6 +862,9 @@ static int m2mtest_open(struct file *file)
 	ctx->transtime = MEM2MEM_DEF_TRANSTIME;
 	ctx->num_processed = 0;
 
+	ctx->q_data[V4L2_M2M_SRC].fmt = &formats[0];
+	ctx->q_data[V4L2_M2M_DST].fmt = &formats[0];
+
 	ctx->m2m_ctx = v4l2_m2m_ctx_init(dev->m2m_dev, ctx, &queue_init);
 
 	if (IS_ERR(ctx->m2m_ctx)) {
@@ -982,9 +987,6 @@ static int m2mtest_probe(struct platform_device *pdev)
 		goto err_m2m;
 	}
 
-	q_data[V4L2_M2M_SRC].fmt = &formats[0];
-	q_data[V4L2_M2M_DST].fmt = &formats[0];
-
 	return 0;
 
 	v4l2_m2m_release(dev->m2m_dev);
-- 
1.7.10

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/