IB/umad: Fix bit ordering and 32-on-64 problems on big endian systems

The declaration of struct ib_user_mad_reg_req.method_mask[] exported
to userspace was an array of __u32, but the kernel internally treated
it as a bitmap made up of longs.  This makes a difference for 64-bit
big-endian kernels, where numbering the bits in an array of__u32 gives:

    |31.....0|63....31|95....64|127...96|

while numbering the bits in an array of longs gives:

    |63..............0|127............64|

64-bit userspace can handle this by just treating method_mask[] as an
array of longs, but 32-bit userspace is really stuck: the meaning of
the bits in method_mask[] depends on whether the kernel is 32-bit or
64-bit, and there's no sane way for userspace to know that.

Fix this by updating <rdma/ib_user_mad.h> to make it clear that
method_mask[] is an array of longs, and using a compat_ioctl method to
convert to an array of 64-bit longs to handle the 32-on-64 problem.
This fixes the interface description to match existing behavior (so
working binaries continue to work) in almost all situations, and gives
consistent semantics in the case of 32-bit userspace that can run on
either a 32-bit or 64-bit kernel, so that the same binary can work for
both 32-on-32 and 32-on-64 systems.
Signed-off-by: Roland Dreier <rolandd@cisco.com>

drivers/infiniband/core/user_mad.c

@@ -44,6 +44,7 @@
 #include <linux/poll.h>
 #include <linux/rwsem.h>
 #include <linux/kref.h>
+#include <linux/compat.h>
 
 #include <asm/uaccess.h>
 #include <asm/semaphore.h>
@@ -607,7 +608,8 @@ static unsigned int ib_umad_poll(struct file *filp, struct poll_table_struct *wa
 	return mask;
 }
 
-static int ib_umad_reg_agent(struct ib_umad_file *file, unsigned long arg)
+static int ib_umad_reg_agent(struct ib_umad_file *file, void __user *arg,
+			     int compat_method_mask)
 {
 	struct ib_user_mad_reg_req ureq;
 	struct ib_mad_reg_req req;
@@ -622,7 +624,7 @@ static int ib_umad_reg_agent(struct ib_umad_file *file, unsigned long arg)
 		goto out;
 	}
 
-	if (copy_from_user(&ureq, (void __user *) arg, sizeof ureq)) {
+	if (copy_from_user(&ureq, arg, sizeof ureq)) {
 		ret = -EFAULT;
 		goto out;
 	}
@@ -643,8 +645,18 @@ found:
 	if (ureq.mgmt_class) {
 		req.mgmt_class         = ureq.mgmt_class;
 		req.mgmt_class_version = ureq.mgmt_class_version;
-		memcpy(req.method_mask, ureq.method_mask, sizeof req.method_mask);
 		memcpy(req.oui, ureq.oui, sizeof req.oui);
+
+		if (compat_method_mask) {
+			u32 *umm = (u32 *) ureq.method_mask;
+			int i;
+
+			for (i = 0; i < BITS_TO_LONGS(IB_MGMT_MAX_METHODS); ++i)
+				req.method_mask[i] =
+					umm[i * 2] | ((u64) umm[i * 2 + 1] << 32);
+		} else
+			memcpy(req.method_mask, ureq.method_mask,
+			       sizeof req.method_mask);
 	}
 
 	agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
@@ -682,13 +694,13 @@ out:
 	return ret;
 }
 
-static int ib_umad_unreg_agent(struct ib_umad_file *file, unsigned long arg)
+static int ib_umad_unreg_agent(struct ib_umad_file *file, u32 __user *arg)
 {
 	struct ib_mad_agent *agent = NULL;
 	u32 id;
 	int ret = 0;
 
-	if (get_user(id, (u32 __user *) arg))
+	if (get_user(id, arg))
 		return -EFAULT;
 
 	down_write(&file->port->mutex);
@@ -729,15 +741,32 @@ static long ib_umad_ioctl(struct file *filp, unsigned int cmd,
 {
 	switch (cmd) {
 	case IB_USER_MAD_REGISTER_AGENT:
-		return ib_umad_reg_agent(filp->private_data, arg);
+		return ib_umad_reg_agent(filp->private_data, (void __user *) arg, 0);
+	case IB_USER_MAD_UNREGISTER_AGENT:
+		return ib_umad_unreg_agent(filp->private_data, (__u32 __user *) arg);
+	case IB_USER_MAD_ENABLE_PKEY:
+		return ib_umad_enable_pkey(filp->private_data);
+	default:
+		return -ENOIOCTLCMD;
+	}
+}
+
+#ifdef CONFIG_COMPAT
+static long ib_umad_compat_ioctl(struct file *filp, unsigned int cmd,
+				 unsigned long arg)
+{
+	switch (cmd) {
+	case IB_USER_MAD_REGISTER_AGENT:
+		return ib_umad_reg_agent(filp->private_data, compat_ptr(arg), 1);
 	case IB_USER_MAD_UNREGISTER_AGENT:
-		return ib_umad_unreg_agent(filp->private_data, arg);
+		return ib_umad_unreg_agent(filp->private_data, compat_ptr(arg));
 	case IB_USER_MAD_ENABLE_PKEY:
 		return ib_umad_enable_pkey(filp->private_data);
 	default:
 		return -ENOIOCTLCMD;
 	}
 }
+#endif
 
 static int ib_umad_open(struct inode *inode, struct file *filp)
 {
@@ -826,7 +855,9 @@ static const struct file_operations umad_fops = {
 	.write 	 	= ib_umad_write,
 	.poll 	 	= ib_umad_poll,
 	.unlocked_ioctl = ib_umad_ioctl,
-	.compat_ioctl 	= ib_umad_ioctl,
+#ifdef CONFIG_COMPAT
+	.compat_ioctl 	= ib_umad_compat_ioctl,
+#endif
 	.open 	 	= ib_umad_open,
 	.release 	= ib_umad_close
 };

include/rdma/ib_user_mad.h

@@ -147,6 +147,26 @@ struct ib_user_mad {
 	__u64	data[0];
 };
 
+/*
+ * Earlier versions of this interface definition declared the
+ * method_mask[] member as an array of __u32 but treated it as a
+ * bitmap made up of longs in the kernel.  This ambiguity meant that
+ * 32-bit big-endian applications that can run on both 32-bit and
+ * 64-bit kernels had no consistent ABI to rely on, and 64-bit
+ * big-endian applications that treated method_mask as being made up
+ * of 32-bit words would have their bitmap misinterpreted.
+ *
+ * To clear up this confusion, we change the declaration of
+ * method_mask[] to use unsigned long and handle the conversion from
+ * 32-bit userspace to 64-bit kernel for big-endian systems in the
+ * compat_ioctl method.  Unfortunately, to keep the structure layout
+ * the same, we need the method_mask[] array to be aligned only to 4
+ * bytes even when long is 64 bits, which forces us into this ugly
+ * typedef.
+ */
+typedef unsigned long __attribute__((aligned(4))) packed_ulong;
+#define IB_USER_MAD_LONGS_PER_METHOD_MASK (128 / (8 * sizeof (long)))
+
 /**
  * ib_user_mad_reg_req - MAD registration request
  * @id - Set by the kernel; used to identify agent in future requests.
@@ -165,7 +185,7 @@ struct ib_user_mad {
  */
 struct ib_user_mad_reg_req {
 	__u32	id;
-	__u32	method_mask[4];
+	packed_ulong method_mask[IB_USER_MAD_LONGS_PER_METHOD_MASK];
 	__u8	qpn;
 	__u8	mgmt_class;
 	__u8	mgmt_class_version;