Skip to contentSkip to conversation

dma/coh901318: fix copy_to_user error path

PATCH In review Nicolas Kaiser 2 participants 4 messages Oct 22, 2010 dmaengine

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 dma/coh901318: fix copy_to_user error path

If copy_to_user fails, the assigned error code instantly gets overwritten, and the failure apparently ignored. Moreover, shouldn't the error code be -EFAULT instead of -EINVAL?

---
 drivers/dma/coh901318.c |    6 ++++--
 1 files changed, 4 insertions(+), 2 deletions(-)
drivers/dma/coh901318.c +4 -2
index ae2b871..eac8867 100644
@@ -140,8 +140,10 @@ static int coh901318_debugfs_read(struct file *file, char __user *buf,
 	if (count > dev_size - *f_pos)
 		count = dev_size - *f_pos;
 
-	if (copy_to_user(buf, dev_buf + *f_pos, count))
-		ret = -EINVAL;
+	if (copy_to_user(buf, dev_buf + *f_pos, count)) {
+		ret = -EFAULT;
+		goto out;
+	}
 	ret = count;
 	*f_pos += count;
 
Signature · 1 line
1.7.2.2
#1 lore raw
Quoted 3 lines · On Friday 22 October 2010 19:29:53 Nicolas Kaiser wrote:

If copy_to_user fails, the assigned error code instantly gets overwritten, and the failure apparently ignored. Moreover, shouldn't the error code be -EFAULT instead of -EINVAL?

Looks good, but it would be even better to just use simple_read_from_buffer in this function, which takes care of a lot the other complexities as well.

Arnd

#2 lore raw

PATCH 1/1 dma/coh901318: use simple_read_from_buffer

* Arnd Bergmann <[email protected]>:

Quoted 8 lines · On Friday 22 October 2010 19:29:53 Nicolas Kaiser wrote:
Quoted 3 lines

If copy_to_user fails, the assigned error code instantly gets overwritten, and the failure apparently ignored. Moreover, shouldn't the error code be -EFAULT instead of -EINVAL?

Looks good, but it would be even better to just use simple_read_from_buffer in this function, which takes care of a lot the other complexities as well.

Like this? (Sorry, untested.)

---
obsoletes patch "dma/coh901318: fix copy_to_user error path"

 drivers/dma/coh901318.c |   12 +-----------
 1 files changed, 1 insertions(+), 11 deletions(-)
drivers/dma/coh901318.c +1 -11
index ae2b871..e7d0f06 100644
@@ -133,17 +133,7 @@ static int coh901318_debugfs_read(struct file *file, char __user *buf,
 	tmp += sprintf(tmp, "Pool alloc nbr %d\n", pool_count);
 	dev_size = tmp  - dev_buf;
 
-	/* No more to read if offset != 0 */
-	if (*f_pos > dev_size)
-		goto out;
-
-	if (count > dev_size - *f_pos)
-		count = dev_size - *f_pos;
-
-	if (copy_to_user(buf, dev_buf + *f_pos, count))
-		ret = -EINVAL;
-	ret = count;
-	*f_pos += count;
+	ret = simple_read_from_buffer(buf, count, f_pos, dev_buf, dev_size);
 
  out:
 	kfree(dev_buf);
Signature · 1 line
1.7.2.2
#3 lore raw

Re: [PATCH] dma/coh901318: use simple_read_from_buffer

Quoted 6 lines · On Friday 22 October 2010 20:30:39 Nicolas Kaiser wrote:

* Arnd Bergmann <[email protected]>:

Quoted 5 lines · On Friday 22 October 2010 19:29:53 Nicolas Kaiser wrote:
Quoted 3 lines

If copy_to_user fails, the assigned error code instantly gets overwritten, and the failure apparently ignored. Moreover, shouldn't the error code be -EFAULT instead of -EINVAL?

Quoted 7 lines
Quoted 3 lines

Looks good, but it would be even better to just use simple_read_from_buffer in this function, which takes care of a lot the other complexities as well.

Like this? (Sorry, untested.)

Yes, exactly. For stable kernels, your smaller fix is probably good enough though and it's obviously correct.

Arnd