Re: KMSAN: uninit-value in _copy_to_iter (2)

On Thu, Jun 07, 2018 at 08:59:06PM +0300, Michael S. Tsirkin wrote:

On Thu, Jun 07, 2018 at 06:43:55PM +0100, Al Viro wrote:

quoted

On Thu, Jun 07, 2018 at 06:38:48PM +0300, Michael S. Tsirkin wrote:

quoted

#syz test: https://github.com/google/kmsan.git/master d2d741e5d1898dfde1a75ea3d29a9a3e2edf0617

Subject: vhost: fix info leak

Fixes: CVE-2018-1118
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
---

diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c
index f0be5f35ab28..9beefa6ed1ce 100644
--- a/drivers/vhost/vhost.c
+++ b/drivers/vhost/vhost.c

@@ -2345,6 +2345,9 @@ struct vhost_msg_node *vhost_new_msg(struct vhost_virtqueue *vq, int type)
 	struct vhost_msg_node *node = kmalloc(sizeof *node, GFP_KERNEL);
 	if (!node)
 		return NULL;
+
+	/* Make sure all padding within the structure is initialized. */
+	memset(&node->msg, 0, sizeof node->msg);

Umm...  Maybe kzalloc(), then?  You have

struct vhost_msg_node {
  struct vhost_msg msg;
  struct vhost_virtqueue *vq;
  struct list_head node;
};

and that's what, 68 bytes in msg, then either 4 bytes pointer or
4 bytes padding + 8 bytes pointer, then two pointers?  How much
does explicit partial memset() save you here?

Yes but 0 isn't a nop here so if this struct is used without
a sensible initialization, it will crash elsewhere.
I prefer KASAN to catch such uses.

quoted

quoted

 	node->vq = vq;
 	node->msg.type = type;

IDGI - what would your variant catch that kzalloc + 2 assignments won't?
Accesses to uninitialized ->node?  Because that's the only difference in
what is and is not initialized between those variants...