[PATCH for-next] RDMA/core: Fix storing node_desc

[PATCH for-next] RDMA/core: Fix storing node_desc

[Kamal Heib]

When writing to node_desc sysfs using echo a new line symbol will be
stored at the end of the string, avoid that by dropping the new line
symbol and also make sure to return -EINVAL when the supplied string is
bigger then IB_DEVICE_NODE_DESC_MAX.

Fixes: c5bcbbb9fe00 ("IB: Allow userspace to set node description")
Signed-off-by: Kamal Heib <kamalheib1@gmail.com>
---
 drivers/infiniband/core/sysfs.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)

diff --git a/drivers/infiniband/core/sysfs.c b/drivers/infiniband/core/sysfs.c
index 087682e6969e..2de5f6710c0b 100644
--- a/drivers/infiniband/core/sysfs.c
+++ b/drivers/infiniband/core/sysfs.c
@@ -1263,12 +1263,21 @@ static ssize_t node_desc_store(struct device *device,
 {
 	struct ib_device *dev = rdma_device_to_ibdev(device);
 	struct ib_device_modify desc = {};
+	size_t len;
 	int ret;
 
 	if (!dev->ops.modify_device)
 		return -EOPNOTSUPP;
 
-	memcpy(desc.node_desc, buf, min_t(int, count, IB_DEVICE_NODE_DESC_MAX));
+	if (count > IB_DEVICE_NODE_DESC_MAX)
+		return -EINVAL;
+
+	len = strlen(buf);
+	if (buf[len - 1] == '\n')
+		len--;
+
+	strncpy(desc.node_desc, buf, len);
+	desc.node_desc[len] = 0;
 	ret = ib_modify_device(dev, IB_DEVICE_MODIFY_NODE_DESC, &desc);
 	if (ret)
 		return ret;
-- 
2.21.0

Re: [PATCH for-next] RDMA/core: Fix storing node_desc

[Jason Gunthorpe]

On Mon, Dec 23, 2019 at 11:39:43AM +0200, Kamal Heib wrote:
> When writing to node_desc sysfs using echo a new line symbol will be
> stored at the end of the string, avoid that by dropping the new line

Why do we want to do this? AFAIK technically new line is valid in a
node description.

> symbol and also make sure to return -EINVAL when the supplied string is
> bigger then IB_DEVICE_NODE_DESC_MAX.

This makes sense though

> diff --git a/drivers/infiniband/core/sysfs.c b/drivers/infiniband/core/sysfs.c
> index 087682e6969e..2de5f6710c0b 100644
> +++ b/drivers/infiniband/core/sysfs.c
> @@ -1263,12 +1263,21 @@ static ssize_t node_desc_store(struct device *device,
>  {
>  	struct ib_device *dev = rdma_device_to_ibdev(device);
>  	struct ib_device_modify desc = {};
> +	size_t len;
>  	int ret;
>  
>  	if (!dev->ops.modify_device)
>  		return -EOPNOTSUPP;
>  
> -	memcpy(desc.node_desc, buf, min_t(int, count, IB_DEVICE_NODE_DESC_MAX));

> +	if (count > IB_DEVICE_NODE_DESC_MAX)
> +		return -EINVAL;
> +
> +	len = strlen(buf);

Why strlen? The buf is count bytes long.

> +	if (buf[len - 1] == '\n')
> +		len--;

And if it is zero bytes this buffer underflows

> +	strncpy(desc.node_desc, buf, len);

What was the point of switching away from memcpy?

> +	desc.node_desc[len] = 0;
>  	ret = ib_modify_device(dev, IB_DEVICE_MODIFY_NODE_DESC, &desc);
>  	if (ret)
>  		return ret;

Jason

Re: [PATCH for-next] RDMA/core: Fix storing node_desc

[Kamal Heib]

On Fri, Jan 03, 2020 at 07:12:12PM -0400, Jason Gunthorpe wrote:
> On Mon, Dec 23, 2019 at 11:39:43AM +0200, Kamal Heib wrote:
> > When writing to node_desc sysfs using echo a new line symbol will be
> > stored at the end of the string, avoid that by dropping the new line
> 
> Why do we want to do this? AFAIK technically new line is valid in a
> node description.
>

Self-Nack, please drop this patch, I didn't do a good job here.

Thanks,
Kamal

> > symbol and also make sure to return -EINVAL when the supplied string is
> > bigger then IB_DEVICE_NODE_DESC_MAX.
> 
> This makes sense though
> 
> > diff --git a/drivers/infiniband/core/sysfs.c b/drivers/infiniband/core/sysfs.c
> > index 087682e6969e..2de5f6710c0b 100644
> > +++ b/drivers/infiniband/core/sysfs.c
> > @@ -1263,12 +1263,21 @@ static ssize_t node_desc_store(struct device *device,
> >  {
> >  	struct ib_device *dev = rdma_device_to_ibdev(device);
> >  	struct ib_device_modify desc = {};
> > +	size_t len;
> >  	int ret;
> >  
> >  	if (!dev->ops.modify_device)
> >  		return -EOPNOTSUPP;
> >  
> > -	memcpy(desc.node_desc, buf, min_t(int, count, IB_DEVICE_NODE_DESC_MAX));
> 
> > +	if (count > IB_DEVICE_NODE_DESC_MAX)
> > +		return -EINVAL;
> > +
> > +	len = strlen(buf);
> 
> Why strlen? The buf is count bytes long.
> 
> > +	if (buf[len - 1] == '\n')
> > +		len--;
> 
> And if it is zero bytes this buffer underflows
> 
> > +	strncpy(desc.node_desc, buf, len);
> 
> What was the point of switching away from memcpy?
> 
> > +	desc.node_desc[len] = 0;
> >  	ret = ib_modify_device(dev, IB_DEVICE_MODIFY_NODE_DESC, &desc);
> >  	if (ret)
> >  		return ret;
> 
> Jason