[PATCH] staging: goldfish: fix direct copy_to_user() from __iomem

[PATCH] staging: goldfish: fix direct copy_to_user() from __iomem

[James A Shackleford]

This patch allocates a few pages and performs an ioread8_rep() from the bus
address, which are then copied to userspace.  This fixes the sparse warning:

drivers/staging/goldfish/goldfish_audio.c:136:43: warning: incorrect type in argument 2 (different address spaces)
drivers/staging/goldfish/goldfish_audio.c:136:43:    expected void const *from
drivers/staging/goldfish/goldfish_audio.c:136:43:    got char [noderef] <asn:2>*read_buffer

which was a result of performing a copy_to_user() directly from the bus address
to the userspace, which can be unsafe across some architectures.

Signed-off-by: James A Shackleford <shack@linux.com>
---
 drivers/staging/goldfish/goldfish_audio.c |   19 +++++++++++++++----
 1 file changed, 15 insertions(+), 4 deletions(-)

diff --git a/drivers/staging/goldfish/goldfish_audio.c b/drivers/staging/goldfish/goldfish_audio.c
index a166424..535ed20 100644
--- a/drivers/staging/goldfish/goldfish_audio.c
+++ b/drivers/staging/goldfish/goldfish_audio.c
@@ -118,10 +118,17 @@ static ssize_t goldfish_audio_read(struct file *fp, char __user *buf,
 	struct goldfish_audio *data = fp->private_data;
 	int length;
 	int result = 0;
+	unsigned int order;
+	void *read_buffer;
 
 	if (!data->read_supported)
 		return -ENODEV;
 
+	order = get_order(READ_BUFFER_SIZE);
+	read_buffer = (void *)__get_free_pages(GFP_KERNEL, order);
+	if (!read_buffer)
+		return -ENOMEM;
+
 	while (count > 0) {
 		length = (count > READ_BUFFER_SIZE ? READ_BUFFER_SIZE : count);
 		AUDIO_WRITE(data, AUDIO_START_READ, length);
@@ -129,17 +136,21 @@ static ssize_t goldfish_audio_read(struct file *fp, char __user *buf,
 		wait_event_interruptible(data->wait,
 			(data->buffer_status & AUDIO_INT_READ_BUFFER_FULL));
 
-		length = AUDIO_READ(data,
-						AUDIO_READ_BUFFER_AVAILABLE);
+		length = AUDIO_READ(data, AUDIO_READ_BUFFER_AVAILABLE);
 
 		/* copy data to user space */
-		if (copy_to_user(buf, data->read_buffer, length))
-			return -EFAULT;
+		ioread8_rep(data->read_buffer, read_buffer, length);
+		if (copy_to_user(buf, read_buffer, length)) {
+			result = -EFAULT;
+			goto error;
+		}
 
 		result += length;
 		buf += length;
 		count -= length;
 	}
+error:
+	__free_pages(read_buffer, order);
 	return result;
 }
 
-- 
1.7.9.5

Re: [PATCH] staging: goldfish: fix direct copy_to_user() from __iomem

[Alan Cox]

On Sun, 2014-06-22 at 03:29 -0400, James A Shackleford wrote:
> This patch allocates a few pages and performs an ioread8_rep() from the bus
> address, which are then copied to userspace.  This fixes the sparse warning:
> 
> drivers/staging/goldfish/goldfish_audio.c:136:43: warning: incorrect type in argument 2 (different address spaces)
> drivers/staging/goldfish/goldfish_audio.c:136:43:    expected void const *from
> drivers/staging/goldfish/goldfish_audio.c:136:43:    got char [noderef] <asn:2>*read_buffer
> 
> which was a result of performing a copy_to_user() directly from the bus address
> to the userspace, which can be unsafe across some architectures.

Goldfish is a specific architecture. It shouldn't be doing direct
xcopies like that - which is why the code is in staging but at the same
time allocating a buffer each call and doing extra copies is not the
right answer.

It should be mapping the pages when they are first set up.

Alan