Re: [PATCH] Reset file->f_op in snd_card_file_remove(). Take 2

[Takashi Iwai <tiwai@suse.de>]

At Wed, 4 Oct 2006 17:36:37 +0200,
Karsten Wiese wrote:
> 
> Am Mittwoch, 4. Oktober 2006 16:18 schrieb Takashi Iwai:
> > At Wed, 4 Oct 2006 12:47:19 +0200,
> > Karsten Wiese wrote:
> > > 
> > > Am Mittwoch, 4. Oktober 2006 11:22 schrieb Takashi Iwai:
> > > > 
> > > > It should call snd_card_free_when_closed() instead.
> > > > 
> > > IMHO, that would just make sure that the bug happens.
> > > Please see my annotations, starting with // in:
> > > 
> > > void fastcall __fput(struct file *file)
> > > {
> > > 	struct dentry *dentry = file->f_dentry;
> > > 	struct vfsmount *mnt = file->f_vfsmnt;
> > > 	struct inode *inode = dentry->d_inode;
> > > 
> > > 	might_sleep();
> > > 
> > > 	fsnotify_close(file);
> > > 	/*
> > > 	 * The function eventpoll_release() should be the first called
> > > 	 * in the file cleanup chain.
> > > 	 */
> > > 	eventpoll_release(file);
> > > 	locks_remove_flock(file);
> > > 
> > > 	if (file->f_op && file->f_op->release)
> > > 		file->f_op->release(inode, file);
> > > // Here snd_hwdep_release() is called.
> > > // snd_hwdep_release() calls snd_card_file_remove().
> > > // snd_card_file_remove() sees card->free_on_last_close ist set,
> > > // calls snd_card_do_free().
> > > // snd_card_do_free frees file->f_op but doesn't set it NULL.
> > > //
> > > 	security_file_free(file);
> > > 	if (unlikely(inode->i_cdev != NULL))
> > > 		cdev_put(inode->i_cdev);
> > > 	fops_put(file->f_op);
> > > // file->f_op has already been freeed!
> > > // fops_put(file->f_op) is likely to oops.
> > 
> > Yes, this looks like an invalid access.
> > 
> > The problem is that we use kmalloc for allocating a dummy f_op.
> > IMO, the simlest solution is to use a static dummy f_op.
> > 
> That'd take 1 static dummy f_op per snd_*_release().

Yes, but it'll remove extra codes at the same time, too.
Well, OTOH, it requires more additions for assignment of dummy ops...

> I prefer the patch at the start of this thread :-)

I think it's not good to set NULL always there.  The NULL is necessary
only when the card is freed.  So, I prefer the patch like below.  Is
it OK?


Takashi

diff -r f38b12373137 sound/core/init.c
--- a/sound/core/init.c	Wed Oct 04 17:17:32 2006 +0200
+++ b/sound/core/init.c	Wed Oct 04 17:50:11 2006 +0200
@@ -721,8 +721,14 @@ int snd_card_file_remove(struct snd_card
 	spin_unlock(&card->files_lock);
 	if (last_close) {
 		wake_up(&card->shutdown_sleep);
-		if (card->free_on_last_close)
+		if (card->free_on_last_close) {
+			/* release and clear f_op here since the dummy f_ops will
+			 * be freed in snd_card_do_free().
+			 */
+			fops_put(file->f_op);
+			file->f_op = NULL;
 			snd_card_do_free(card);
+		}
 	}
 	if (!mfile) {
 		snd_printk(KERN_ERR "ALSA card file remove problem (%p)\n", file);