Re: access_ok inside kernelspace

Tigran Aivazian (tigran@sco.COM)
Tue, 19 Oct 1999 07:59:54 +0100 (BST)


Hi Alan (and the World),

You said recently:
----------------------------- begin >
mm_segment_t fs;

fs=get_fs();
set_fs(get_ds());

/* User space is now the kernel */

/* Put it back again */

set_fs(fs);

------------------------------- end >

However, it seems to be not *always* safe. Please have a look at this
tiny piece of code, if you have 2 minutes:

int __init init_bfs_fs(void)
{
int error;
mm_segment_t fs;

error = register_filesystem(&bfs_fs_type);
if (error)
return error;
fs = get_fs();
set_fs(get_ds());
bfs_fsindex = sysfs(1, (unsigned long)"bfs", 0);
set_fs(fs);
return bfs_fsindex < 0 ? -EINVAL : 0;
}

(where sysfs() is pulled in via _syscall3() from <asm/unistd.h> and errno
is exported and __KERNEL_SYSCALLS__ is defined, of course).

It works fine if the filesystem is already loaded (insmod bfs or
statically linked) and then mounted but oopses (in request_module() from
get_fs_type()) if it is autoloaded from within mount(2) system call.

So, the moral (temporary moral, until I see the picture clearer) of the
story is: do not use system calls from within system calls. Is this
correct or is it a bug that I should try to fix?

If the above behaviour is correct then I would appreciate anyone
suggesting some ideas on how to get at fs_index() value without exporting
fs_index() and without doing the syscall trick above. (I abandon my
earlier idea of exporting fs_index() because I believe all accesses to
file_systems should go through rw spinlock and it (nor fs_name nor
fs_maxindex) shouldn't be exported until that is done).

Al Viro suggested adding a field to struct file_system_type which is
initialized by register_filesystem(), which seems to be a good idea. If
no-one suggests anything better I will do so. (I *really* need fs_index()
value - pretty please :)

Regards,
------
Tigran A. Aivazian | http://www.sco.com
Escalations Research Group | tel: +44-(0)1923-813796
Santa Cruz Operation Ltd | http://www.ocston.org/~tigran

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/