Re: accessing file operations directly in the kernel

Re: accessing file operations directly in the kernel

[Robert Hancock]

andy wrote:
> Hi,
> 
> I am dealing with an interesting situation, which I have to access a
> serial device driver
> from the kernel space. I tried to use sys_open, sys_ioctl ... but
> there is a problem
> with the file descriptors.  Now, I started using direct calls to the
> drivers functions
> driver_open, driver_ioctl... passing the *inode from user_path_walk(),
> but it doesn't
> seem that works right. This serial device is going to be used only
> from the other
> module that runs in the kernel.

For some reason this kind of question (using file operations inside the 
kernel) gets asked a lot. The short answer is that if you're asking it, 
you likely have a wrong design. It's just not the right thing to do. For 
one thing, file descriptors belong to a process, so just stealing the FD 
space from some random process is unlikely to do the right thing.

If you really need to do this, adding hooks to the serial device driver 
would seem a more logical thing to do.. but even that seems a suspicious 
design. It would be best if you explained what it is you're actually 
trying to accomplish.

Re: accessing file operations directly in the kernel

[andy]

I agree with you! But there are cases where standards
are followed from certain developers. I have to deal with a
very "peculiar" arrangement of a set of requirements to access
a serial device driver. The type of code below gives me what I need, I think.

filp = filp_open("/dev/device1", O_WRONLY, 0);

if (filp->f_op->ioctl(filp->f_dentry->d_inode, filp,
							CMD,
							(unsigned long) &args) != 0) {
        		goto failed;
}

Regards
Andy

On 11/7/08, Robert Hancock <hancockr@shaw.ca> wrote:
> andy wrote:
>
> > Hi,
> >
> > I am dealing with an interesting situation, which I have to access a
> > serial device driver
> > from the kernel space. I tried to use sys_open, sys_ioctl ... but
> > there is a problem
> > with the file descriptors.  Now, I started using direct calls to the
> > drivers functions
> > driver_open, driver_ioctl... passing the *inode from user_path_walk(),
> > but it doesn't
> > seem that works right. This serial device is going to be used only
> > from the other
> > module that runs in the kernel.
> >
>
>  For some reason this kind of question (using file operations inside the
> kernel) gets asked a lot. The short answer is that if you're asking it, you
> likely have a wrong design. It's just not the right thing to do. For one
> thing, file descriptors belong to a process, so just stealing the FD space
> from some random process is unlikely to do the right thing.
>
>  If you really need to do this, adding hooks to the serial device driver
> would seem a more logical thing to do.. but even that seems a suspicious
> design. It would be best if you explained what it is you're actually trying
> to accomplish.
>

accessing file operations directly in the kernel

[andy]

Hi,

I am dealing with an interesting situation, which I have to access a
serial device driver
from the kernel space. I tried to use sys_open, sys_ioctl ... but
there is a problem
with the file descriptors.  Now, I started using direct calls to the
drivers functions
driver_open, driver_ioctl... passing the *inode from user_path_walk(),
but it doesn't
seem that works right. This serial device is going to be used only
from the other
module that runs in the kernel.

TIA
Andy