Re: [PATCH 06/19] Add ioctl() argument and attribute handling utils

Miloslav Trmac wrote:

----- "Stefan Richter" [off-list ref] wrote:

quoted

Miloslav Trmač wrote:

quoted

--- /dev/null
+++ b/crypto/userspace/utils.c

[...]

quoted

+#ifdef CONFIG_COMPAT
+/* max() is too clever for compile-time constants */
+#define CONST_MAX(A, B) ((A) > (B) ? (A) : (B))
+
+#define MAX_SESSION_INPUT_DATA_SIZE					\
+	(CONST_MAX(sizeof(struct ncr_session_input_data),		\
+		   sizeof(struct compat_ncr_session_input_data)))
+#define MAX_SESSION_OUTPUT_BUFFER_SIZE					\
+	(CONST_MAX(sizeof(struct ncr_session_output_buffer),		\
+		   sizeof(struct compat_ncr_session_output_buffer)))
+
+#else /* !CONFIG_COMPAT */
+
+#define MAX_SESSION_INPUT_DATA_SIZE (sizeof(struct ncr_session_input_data))
+#define MAX_SESSION_OUTPUT_BUFFER_SIZE			\
+	(sizeof(struct ncr_session_output_buffer))
+
+#endif /* !CONFIG_COMPAT */

struct compat_xyz cannot be bigger than struct xyz, can it?

Perhaps not, but the code has been written and it's easier to keep it than to
speculate :)

What's there to speculate? :-)  COMPAT = 32bit user on 64bit kernel.  The
latter has larger long and pointers and possibly different alignment of 64bit
data.

It also helps document the fact that the data format depends on properties of
the userspace.

Since it is a new ABI, it can be written to not need COMPAT ifdef'ery nor
compat handlers.
-- 
Stefan Richter
-=====-==-=- =--- =-=-=
http://arcgraph.de/sr/