ethtool: make .get_dump_data() harder to misuse by drivers

As the patch "bnx2x: remove zeroing of dump data buffer" showed,
it is too easy implement .get_dump_data incorrectly in a driver.

Let's make sure drivers cannot get confused by userspace requesting
a too big dump.

Also WARN if the driver sets dump->len to something weird and make
sure the length reported to userspace is the actual length of data
copied to userspace.

Signed-off-by: Michal Schmidt <mschmidt@redhat.com>
Reviewed-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/core/ethtool.c

@@ -1320,10 +1320,19 @@ static int ethtool_get_dump_data(struct net_device *dev,
 	if (ret)
 		return ret;
 
-	len = (tmp.len > dump.len) ? dump.len : tmp.len;
+	len = min(tmp.len, dump.len);
 	if (!len)
 		return -EFAULT;
 
+	/* Don't ever let the driver think there's more space available
+	 * than it requested with .get_dump_flag().
+	 */
+	dump.len = len;
+
+	/* Always allocate enough space to hold the whole thing so that the
+	 * driver does not need to check the length and bother with partial
+	 * dumping.
+	 */
 	data = vzalloc(tmp.len);
 	if (!data)
 		return -ENOMEM;
@@ -1331,6 +1340,16 @@ static int ethtool_get_dump_data(struct net_device *dev,
 	if (ret)
 		goto out;
 
+	/* There are two sane possibilities:
+	 * 1. The driver's .get_dump_data() does not touch dump.len.
+	 * 2. Or it may set dump.len to how much it really writes, which
+	 *    should be tmp.len (or len if it can do a partial dump).
+	 * In any case respond to userspace with the actual length of data
+	 * it's receiving.
+	 */
+	WARN_ON(dump.len != len && dump.len != tmp.len);
+	dump.len = len;
+
 	if (copy_to_user(useraddr, &dump, sizeof(dump))) {
 		ret = -EFAULT;
 		goto out;