On Tue, Feb 02, 2010 at 12:18:35AM +0100, John Kacur wrote:

Sorry, maybe I am doing a poor job of explaining myself. My question
was whether your driver needs to call uinput_release() or not if it
went through your proposed error path, because that is where you have
the call to the uinput_destroy_device() function.
After taking a fresh look at your code I don't believe that it does.
However, you could still hoist your code that calls nonseekable_open()
above all that init stuff in uinput_open(), just under the return
-ENOMEM if you think that it could fail.
However, I still think that nonseekable_open() is designed from the
"get-go" to never fail, so I think your code is unnecessarily
complicated, by just a little bit. It will still work, so you decide
which to go with. I'm fine with either way.

OK, so how about the patch below? If it is accepted I will just switch
to

	nonseekable_open(inode, file);
	return 0;

style. I gonna add Al and akpm to CC to see if the patch will stick...

-- 
Dmitry

VFS: clarify that nonseekable_open() will never fail

Signed-off-by: Dmitry Torokhov <redacted>
---

 fs/open.c |    4 +++-
 1 files changed, 3 insertions(+), 1 deletions(-)

diff --git a/fs/open.c b/fs/open.c
index 040cef7..02ceb73 100644
--- a/fs/open.c
+++ b/fs/open.c

@@ -1200,7 +1200,9 @@ EXPORT_SYMBOL(generic_file_open);
 
 /*
  * This is used by subsystems that don't want seekable
- * file descriptors
+ * file descriptors. The function is not supposed to ever fail, the only
+ * reason it returns an 'int' and not 'void' is so that it can be plugged
+ * directly into file_operations structure.
  */
 int nonseekable_open(struct inode *inode, struct file *filp)
 {