Skip to contentSkip to conversation

userspace vs. kernelspace address

DISCUSSION Rock Gordon 4 participants 4 messages Jan 28, 2005 – Jan 30, 2005 1 message not in the archive

lore
Share & tools
lore
#1 lore raw
In reply to a message not in the archive

Hi everbody,

Thanks for your replies.

Lemme explain my problem a little bit more .... I have a thread that does exactly similar things in kernel-mode and user-mode (depending on how you invoked it; of course, the kernel one is forked using kernel_thread(), and the user one is from pthread_create()). The architecture-dependant stuff is taken care of by extensive use of __KERNEL__ macro testing.

This particular thread gets a packet of data, the header of which contains address to where it should be copying the payload associated with that packet. The kernel-mode thread will need to decide how to copy data into another process' address space, so will the user-mode thread.

However I think my copy_to_user and copy_from_user are failing since the kernel-mode thread is copying data into another process's address space, and I am not sure how to do this. Do the get_fs() and set_fs() combinations let you do that? If not, then how do I do it?

Something like when you invoke the ->write or ->read functions, you need to copy the requisite data into the buffer the application provided you with.

Thanks and regards,
Rock

Quoted 23 lines · --- Jan Hudec
Quoted 1 line · Bernd Petrovitsch

On Thu, 2005-01-27 at 09:14 -0800, Rock Gordon

Quoted 2 lines · wrote:
Quoted 2 lines

If I'm given a particular address, how do I test
whether that address is from userspace or from

kernel

Quoted 5 lines
Quoted 1 line

space?

You don't.

Quoted 1 line

I need to make these decisions from either

inside a

Quoted 1 line
Quoted 1 line

kernel module or a userspace program. The idea

is I

Quoted 2 lines
Quoted 2 lines

use memcpy() in the user-user version,
copy_from/to_user in the kernel-kernel version,

and

Quoted 3 lines
Quoted 1 line

prohibit the others.

You need to know where the address is from and use

the correct function.

Quoted 9 lines

If the interface is defined as taking userland address, than kernel
function passing a kernel address in is responsible for calling
set_fs(KERNEL_DS) before and undoing it after. That way the
copy_to/from_user does not complain.

-------------------------------------------------------------------------------

Quoted 4 lines

Jan 'Bulb' Hudec <[email protected]>

ATTACHMENT part 2 application/pgp-signature

name=signature.asc

__________________________________
Do you Yahoo!?
Take Yahoo! Mail with you! Get it on your mobile phone.
http://mobile.yahoo.com/maildemo

#2 lore raw
Quoted 9 lines · Rock Gordon

Hi everbody,

Thanks for your replies.

However I think my copy_to_user and copy_from_user are failing since the kernel-mode thread is copying data into another process's address space, and I am not sure how to do this. Do the get_fs() and set_fs() combinations let you do that? If not, then how do I do

My idea is on kernel thread is limited. But I think it is not possible to any userspace address from any kernel thread because they do not have access to it. Their proc_struct->mm field is empty.
I am not sure whether set_fs and get_fs help in this case.

HTH,
Om

#3 lore raw
Quoted 5 lines · Om
Quoted 4 lines · Rock Gordon

Hi everbody,

Thanks for your replies.

Quoted 8 lines
Quoted 5 lines

However I think my copy_to_user and copy_from_user are failing since the kernel-mode thread is copying data into another process's address space, and I am not sure how to do this. Do the get_fs() and set_fs() combinations let you do that? If not, then how do I do

My idea is on kernel thread is limited. But I think it is not possible to any userspace address from any kernel thread because they do not have access to it. Their proc_struct->mm field is empty.

Right. You can't access any user-space from kernel thread, because it does not have any.

Quoted 1 line

I am not sure whether set_fs and get_fs help in this case.

Sure it can. set_fs(KERNEL_DS) sets things so, that if you pass kernel address to copy_to/from_user, it will silently accept it and copy to/from there.

-------------------------------------------------------------------------------
Jan 'Bulb' Hudec <[email protected]>

#4 lore raw
In reply to Rock Gordon
Quoted 8 lines · Rock Gordon

Lemme explain my problem a little bit more .... I have a thread that does exactly similar things in kernel-mode and user-mode (depending on how you invoked it; of course, the kernel one is forked using kernel_thread(), and the user one is from pthread_create()). The architecture-dependant stuff is taken care of by extensive use of __KERNEL__ macro testing.

You can than use the same macros for getting to correct copying function.

Bernd

Signature · 3 lines
Firmix Software GmbH                   http://www.firmix.at/
mobil: +43 664 4416156                 fax: +43 1 7890849-55
          Embedded Linux Development and Services