create a file in kernel mode. help please!

Previous thread: [PATCH for mm] Remove iBCS support by Andi Kleen on Saturday, January 19, 2008 - 12:59 am. (40 messages)

Next thread: [PATCH] kernel/params.c: fix the module name length in param_sysfs_builtin by rae l on Saturday, January 19, 2008 - 1:46 am. (1 message)
To: Linux kernel <linux-kernel@...>
Date: Saturday, January 19, 2008 - 1:17 am

Hi there,
obviously this is a newbie question, but I couldn't find any
documentation on how to do it.. I tried several ways but couldnt do
it.
I designed a system call, so a user will call it, and a new file will
be created ('/tmp/filexx'). After that, I have another system call,
that will map the file into the maps of the user process. The idea is
the same as IPC...

I managed to create the file with this function (in the first system call):
fd = filp_open(path, O_CREAT | O_RDWR , 777);

After that, the user will call another system call, and it will map
this file to the process maps.
something like this:
filp_open(route, O_RDWR,0 );
do_mmap(fp, 0, tamano, PROT_READ | PROT_WRITE | PROT_EXEC,
MAP_SHARED, 0);

After I call the second system call, the user tries to access the
memory, but gets the message "Bus Error".
I tried to manually create a file with vi, and then use the second
system call, and worked perfectly. I could use the shared memory
without problems.

The problems seems to be in the first system call (with filp_open),
when I try to create a new file... Can somebody suggest me something,
on how I could fix this issue?? It is very important because it is for
a college projects.

Greetings, and thanks in advance for the answers.
Rafael Sisto - Uruguay.-
--

To: Rafael Sisto <rafael.sisto@...>
Cc: Linux kernel <linux-kernel@...>
Date: Wednesday, January 23, 2008 - 2:02 pm

Hmm, I don't quite get what you are trying to achieve? Why can't user
create the file via open(path, O_CREAT...) and then mmap() it? Why do
you need a new system call for that?

Honza
--
Jan Kara <jack@suse.cz>
SuSE CR Labs
--

To: Rafael Sisto <rafael.sisto@...>
Cc: Linux kernel <linux-kernel@...>
Date: Monday, January 21, 2008 - 4:21 am

Hi Rafael.

Probably you haven't got any answers yet because you should NEVER EVER
(i repeat: NEVER EVER, not even in college projects) go for file I/O in
the kernel. You might want to think the whole thing over one or two
times, i think.

If it comes to you that it is absolutely necessary (i repeat again:
ABSOLUTELY NECESSARY, like a solution to feed the world) then you might
google for some ancient FAQ of Greg (K-H) (i think he was) where he
tries to give a little more details about this kind of problem, and also
gives a possible solution, IIRC. Sorry that i don't remember the exact
page/adress anymore, for i intend never to look at it again.

Greetz
Joe

--
Sepp "ZaP" Holzmayr

please reply to:
zentrale.at.work@gmail.com

watch out for:
www.rocksociety.de
--

To: Joe <zentrale.at.work@...>
Cc: Linux kernel <linux-kernel@...>
Date: Monday, January 21, 2008 - 8:44 am

Thanks Joe for the answer! I will find another work around for my problem, then!

greetings, Rafael Sisto.

--

Previous thread: [PATCH for mm] Remove iBCS support by Andi Kleen on Saturday, January 19, 2008 - 12:59 am. (40 messages)

Next thread: [PATCH] kernel/params.c: fix the module name length in param_sysfs_builtin by rae l on Saturday, January 19, 2008 - 1:46 am. (1 message)