[PATCH] generic_nvram: use inode lock not lock_kernel

[PATCH] generic_nvram: use inode lock not lock_kernel

[Alan Cox]

The locking is only done for lseek so this isn't a big change. All the
other operations are thrown at the arch specific lower layers for locking
handling without the file pointer being directly exposed.

Signed-off-by: Alan Cox <alan@redhat.com>

diff -u --new-file --recursive --exclude-from /usr/src/exclude linux.vanilla-2.6.22-rc6-mm1/drivers/char/generic_nvram.c linux-2.6.22-rc6-mm1/drivers/char/generic_nvram.c
--- linux.vanilla-2.6.22-rc6-mm1/drivers/char/generic_nvram.c	2007-07-02 20:47:23.000000000 +0100
+++ linux-2.6.22-rc6-mm1/drivers/char/generic_nvram.c	2007-07-09 12:05:06.786669552 +0100
@@ -30,7 +30,7 @@
 
 static loff_t nvram_llseek(struct file *file, loff_t offset, int origin)
 {
-	lock_kernel();
+	mutex_lock(&file->f_path.dentry->d_inode->i_mutex);
 	switch (origin) {
 	case 1:
 		offset += file->f_pos;
@@ -44,7 +44,7 @@
 		return -EINVAL;
 	}
 	file->f_pos = offset;
-	unlock_kernel();
+	mutex_unlock(&file->f_path.dentry->d_inode->i_mutex);
 	return file->f_pos;
 }
 

Re: [PATCH] generic_nvram: use inode lock not lock_kernel

[Andrew Morton]

On Mon, 9 Jul 2007 14:47:45 +0100
Alan Cox <alan@lxorguk.ukuu.org.uk> wrote:

> The locking is only done for lseek so this isn't a big change. All the
> other operations are thrown at the arch specific lower layers for locking
> handling without the file pointer being directly exposed.
> 
> Signed-off-by: Alan Cox <alan@redhat.com>
> 
> diff -u --new-file --recursive --exclude-from /usr/src/exclude linux.vanilla-2.6.22-rc6-mm1/drivers/char/generic_nvram.c linux-2.6.22-rc6-mm1/drivers/char/generic_nvram.c
> --- linux.vanilla-2.6.22-rc6-mm1/drivers/char/generic_nvram.c	2007-07-02 20:47:23.000000000 +0100
> +++ linux-2.6.22-rc6-mm1/drivers/char/generic_nvram.c	2007-07-09 12:05:06.786669552 +0100
> @@ -30,7 +30,7 @@
>  
>  static loff_t nvram_llseek(struct file *file, loff_t offset, int origin)
>  {
> -	lock_kernel();
> +	mutex_lock(&file->f_path.dentry->d_inode->i_mutex);
>  	switch (origin) {
>  	case 1:
>  		offset += file->f_pos;
> @@ -44,7 +44,7 @@
>  		return -EINVAL;
>  	}
>  	file->f_pos = offset;
> -	unlock_kernel();
> +	mutex_unlock(&file->f_path.dentry->d_inode->i_mutex);
>  	return file->f_pos;
>  }
>  

You left an unlock_kernel() on the error path, thus introducing a deadly
bug.

Can we just use generic_file_llseek() in here?  afacit that simply requires
that i_size have the correct value.  Does it?

generic_file_llseek() uses file->f_mapping->host->i_mutex which is
equivalent for this file, but we might as well be consistent.

While we're there, we can do s/magic numbers/standard symbols/.

And we can fix the remaining race in there by reading f_pos while holding
the mutex, not after having dropped it.


End result:


static loff_t nvram_llseek(struct file *file, loff_t offset, int origin)
{
	struct inode *inode = file->f_mapping->host;

	mutex_lock(&inode->i_mutex);
	switch (origin) {
	case SEEK_CUR:
		offset += file->f_pos;
		break;
	case SEEK_END:
		offset += NVRAM_SIZE;
		break;
	}
	if (offset < 0) {
		offset = -EINVAL;
		goto out;
	}
	file->f_pos = offset;
out:
	mutex_unlock(&inode->i_mutex);
	return offset;
}

Re: [PATCH] generic_nvram: use inode lock not lock_kernel

[Alan Cox]

> Can we just use generic_file_llseek() in here?  afacit that simply requires
> that i_size have the correct value.  Does it?

Yes.

> Generic_file_llseek() uses file->f_mapping->host->i_mutex which is
> equivalent for this file, but we might as well be consistent.

Last time I used file->f_mapping->host-> for this in another driver Al
Viro asked me to use the f_path sequence instead. So I've always done
that since.

If it is correct to use the other then generic_file_llseek looks fine, if
not perhaps we need generic_device_llseek() ?

Al - there are many clones of these lseek methods - which should we be
using ?