Re: Question about sys32_ioctl()

Re: Question about sys32_ioctl()

[Thierry Lafage]

Hello,

Nobody answered the question I posted yesterday, so I re-post it now
in case it was taken as a 1st april joke ;-). But this was not, even
if it sounds stupid from me to put my nose in the kernel sources...

The sys32_ioctl() code contains things I really do not understand, but
since I applied the modification I propose below, I would like to know
if it is a bad thing for the kernel stability and/or mechanisms. And
if it is, I apologize for the mailing lists pollution (as I should do
every time I post here).

Thierry Lafage.

------- Start of forwarded message -------

Hello,

I have made a personal char device driver. For the device, I would
like to add new ioctl commands (I have read the file
"Documentation/ioctl-number.txt" for that). But with sparc32 user
programs, the new ioctl commands cannot be serviced by my char driver:

sys32_ioctl: Unknown cmd fd(3) cmd(2000ee00) arg(00000000)

When looking in the code "arch/sparc64/kernel/ioctl32.c", the
sys32_ioctl() function never calls such added cmds: 

asmlinkage int sys32_ioctl(unsigned int fd, unsigned int cmd, unsigned long arg)
{
[...]
        switch (cmd) {
[...]
        default:
                do {
                        static int count = 0;
                        if (++count <= 20)
                                printk("sys32_ioctl: Unknown cmd fd(%d) "
                                       "cmd(%08x) arg(%08x)\n",
                                       (int)fd, (unsigned int)cmd, (unsigned int)arg);
                } while(0);
                error = -EINVAL;
                break;
        }
out:
        unlock_kernel();
        return error;
}


Would it be stupid to add something like what follows to the default
case (inspired from "fs/ioctl.c: sys_ioctl(...)")?

if (filp->f_op && filp->f_op->ioctl && filp->f_dentry && filp->f_dentry->d_inode)
	       error = filp->f_op->ioctl(filp->f_dentry->d_inode, filp, cmd, arg);

Thierry Lafage.
- -
To unsubscribe from this list: send the line "unsubscribe sparclinux" in
the body of the message to majordomo@vger.rutgers.edu
------- End of forwarded message -------

Re: Question about sys32_ioctl()

[David Miller]

   Date: 	Fri, 2 Apr 1999 17:35:54 +0200
   From: Thierry Lafage <lafage@irisa.fr>

   The sys32_ioctl() code contains things I really do not understand,
   but since I applied the modification I propose below, I would like
   to know if it is a bad thing for the kernel stability and/or
   mechanisms.

I might put this in, but I really don't think I will because the core
problem is that this will let through ioctl's which need 32bit-->64bit
translation on their arguments and we'll have a hard time finding them
with the change in there which you propose.

Your driver may be OK, but other people's will surely not.

So just get your driver merged into the kernel, and then I'll know
about your ioctl's and can add them explicitly to the sys32_ioctl()
switch statement and this whole problem is solved.

Later,
David S. Miller
davem@redhat.com