How do I call Linux syscall from kernel space?

Question

I'm porting linux kernel module written for Linux 2.4 to work with Linux 2.6. Some syscalls declared through syscallN() macros and wrapped in set_fs() calls were used in the code. How can I still use sycalls in Linux 2.6 where those macros are absent?

I know it's a bad taste to use syscalls from kernel space and syscallN() macros are broken on most platforms. Any reasonable way to replace getuid, geteuid, mknod, chown, unlink, sched_yield syscalls in kernel space is appreciated.

Answer 1

current->uid and current->euid can substitute for the first two.

schedule() should work for the last one.

The filesystem operations look more complicated: you might try and see if sys_chown(), sys_mknod(), and sys_unlink() are exported (available for use by any module). If they work, great. There are some useful tips here. Otherwise, you have to dig a little deeper:

The chown syscall is defined in fs/open.c. At a glance I don't see why you couldn't copy that code into your own "kernel_chown" function and give it a try.

The mknodat and unlink syscalls are in fs/namei.c; they eventually wind up calling vfs_mknod() and vfs_unlink(), respectively. Maybe you can duplicate that code or figure out how it's done from there.