Re: [PATCH] move common compat ioctls to hash

["Michael S. Tsirkin" <mst@mellanox.co.il>]

Quoting r. Andi Kleen <ak@suse.de>:
> Subject: Re: [PATCH] move common compat ioctls to hash
> 
> On Mon, Jan 24, 2005 at 10:26:09PM +0200, Michael S. Tsirkin wrote:
> > Hi!
> > The new ioctl code in fs/compat.c can be streamlined a little
> > using the compat hash instead of an explicit switch statement.
> > 
> > The attached patch is against 2.6.11-rc2-bk2.
> > Andi, could you please comment? Does this make sence?
> 
> Problem is that when a compat_ioctl handler returns -EINVAL
> instead of -ENOIOCTLCMD on unknown ioctl it won't check the common
> ones.
> 
> I fear this mistake would be common, that is why I put in the switch.
> 
> -Andi

Still, many drivers still need the compat hash lookup to work
properly, and these still require -ENOIOCTLCMD to work properly.
So is it worth it wasting CPU cycles working around only part of the problem?

How about:
1. Adding a comment in fs.h
2. Changing unlocked_ioctl to behave in the same way as compat_ioctl
 (this will help since unlocked_ioctl will be well tested),
 something along the lines of my old patch below [its against 2.6.11-rc1-bk5].

As a bonus, I expect a small speedup in code using unlocked_ioctl
(as well as compat_ioctl) for datapath things like usb transfers.

Andy, if this sounds convincing, let me know and I'll build a real patch.

MST

diff -rup linux-2.6.10-orig/fs/ioctl.c linux-2.6.10-ioctl-sym/fs/ioctl.c
--- linux-2.6.10-orig/fs/ioctl.c	2005-01-18 10:58:33.609880024 +0200
+++ linux-2.6.10-ioctl-sym/fs/ioctl.c	2005-01-18 10:51:55.690372984 +0200
@@ -24,12 +24,7 @@ static long do_ioctl(struct file *filp, 
 	if (!filp->f_op)
 		goto out;
 
-	if (filp->f_op->unlocked_ioctl) {
-		error = filp->f_op->unlocked_ioctl(filp, cmd, arg);
-		if (error == -ENOIOCTLCMD)
-			error = -EINVAL;
-		goto out;
-	} else if (filp->f_op->ioctl) {
+	if (filp->f_op->ioctl) {
 		lock_kernel();
 		error = filp->f_op->ioctl(filp->f_dentry->d_inode,
 					  filp, cmd, arg);
@@ -93,6 +91,12 @@ asmlinkage long sys_ioctl(unsigned int f
  	if (error)
  		goto out_fput;
 
+	if (filp->f_op && filp->f_op->unlocked_ioctl) {
+		error = filp->f_op->unlocked_ioctl(filp, cmd, arg);
+		if (error != -ENOIOCTLCMD)
+			goto out_fput;
+	}
+
 	switch (cmd) {
 		case FIOCLEX:
 			set_close_on_exec(fd, 1);
-- 
I dont speak for Mellanox.