Re: [PATCH] 2.5: push BKL out of llseek

[Robert Love]

On Tue, 2002-01-29 at 19:09, Linus Torvalds wrote:

> Thinking about that, that actally sounds like the _right_ thing to do even
> from a correctness standpoint - as llseek() looks at the inode size, so we
> should have that lock anyway.
> 
> So I'd suggest doing the inode semaphore globally, instead of using
> kernel_lock at all.

This was my end result (or some sort of proper finer-grained lock) but I
was unsure of how to deal with the other llseek methods that may be
oddball.  I have no reason to suspect using the inode semaphore won't
work.

Another gain from pushing the locks into each method is that we can pick
and choose as-needed.  If it turns out inode semaphore is a global
solution, the following patch is sufficient.  Otherwise, we could
replace the lock_kernel in each caller with the inode semaphore, as
appropriate.  Oh Al ??

	Robert Love

--- linux-2.5.3-pre6/fs/read_write.c	Mon Jan 28 18:30:22 2002
+++ linux/fs/read_write.c	Tue Jan 29 19:29:32 2002
@@ -84,9 +84,9 @@
 	fn = default_llseek;
 	if (file->f_op && file->f_op->llseek)
 		fn = file->f_op->llseek;
-	lock_kernel();
+	down(&file->f_dentry->d_inode->i_sem);
 	retval = fn(file, offset, origin);
-	unlock_kernel();
+	up(&file->f_dentry->d_inode->i_sem);
 	return retval;
 }