Re: export of sys_call_table

[Andi Kleen <ak@suse.de>]

On Fri, Oct 04, 2002 at 01:15:47PM -0600, Brian F. G. Bidulock wrote:
> Alan,
> 
> On Fri, 04 Oct 2002, Alan Cox wrote:
> > 
> > AFS patches a collection of random syscalls in pretty icky ways. Again
> > afssyscall wants doing the right way - with a kernel stub like NFS has
> > 
> 
> Attached is an untested patch for LiS.  AK doesn't like the read/write_lock.
> AFAIK it will work for LiS but might not work for AFS.  Also LiS doesn't need
> module load like nfsd, perhaps afs does.  If someone could add an afs patch we
> could kill these two birds with one stone.  Patch is against 2.4.18 but should
> move up fine.
> 
> --brian
> 
>  kernel/ksyms.c                |    2 ++
>  kernel/sys.c                  |   39 +++++++++++++++++++++++++++++++++++++++
>  arch/alpha/kernel/entry.S     |    2 ++
>  arch/arm/kernel/calls.S       |    4 !!!!
>  arch/cris/kernel/entry.S      |    4 !!!!
>  arch/i386/kernel/entry.S      |    4 !!!!
>  arch/ia64/kernel/entry.S      |    4 !!!!
>  arch/m68k/kernel/entry.S      |    4 !!!!
>  arch/mips/kernel/syscalls.h   |    4 !!!!
>  arch/mips64/kernel/scall_64.S |    4 !!!!
>  arch/parisc/kernel/syscall.S  |    4 !!!!
>  arch/ppc/kernel/misc.S        |    4 !!!!
>  arch/s390/kernel/entry.S      |    4 !!!!
>  arch/s390x/kernel/entry.S     |    4 !!!!
>  arch/sh/kernel/entry.S        |    4 !!!!
>  arch/sparc/kernel/systbls.S   |    4 !!!!
>  arch/sparc64/kernel/systbls.S |    6 !!!!!!
>  include/asm-alpha/unistd.h    |    2 ++
>  include/asm-arm/unistd.h      |    4 !!!!
>  include/asm-sh/unistd.h       |    4 !!!!
>  include/asm-sparc/unistd.h    |    4 !!!!
>  include/asm-sparc64/unistd.h  |    4 !!!!
>  22 files changed, 45 insertions(+), 74 modifications(!)
> 
> 
> Index: kernel/ksyms.c
> ===================================================================
> RCS file: /home/common/cvsroot/linux/kernel/ksyms.c,v
> retrieving revision 1.1.3.1
> diff -c -r1.1.3.1 ksyms.c
> *** kernel/ksyms.c	25 Feb 2002 19:38:13 -0000	1.1.3.1
> --- kernel/ksyms.c	4 Oct 2002 17:17:09 -0000
> ***************
> *** 469,474 ****
> --- 469,476 ----
>   #ifndef __mips__
>   EXPORT_SYMBOL(sys_call_table);
>   #endif
> + EXPORT_SYMBOL(register_streams_calls);
> + EXPORT_SYMBOL(unregister_streams_calls);
>   EXPORT_SYMBOL(machine_restart);
>   EXPORT_SYMBOL(machine_halt);
>   EXPORT_SYMBOL(machine_power_off);
> Index: kernel/sys.c
> ===================================================================
> RCS file: /home/common/cvsroot/linux/kernel/sys.c,v
> retrieving revision 1.1.3.1
> diff -c -r1.1.3.1 sys.c
> *** kernel/sys.c	25 Feb 2002 19:38:13 -0000	1.1.3.1
> --- kernel/sys.c	4 Oct 2002 18:45:03 -0000
> ***************
> *** 167,172 ****
> --- 167,211 ----
>   	return notifier_chain_unregister(&reboot_notifier_list, nb);
>   }
>   
> + static int (*do_putpmsg) (int, void *, void *, int, int) = NULL;
> + static int (*do_getpmsg) (int, void *, void *, int, int) = NULL;
> + 
> + static rwlock_t streams_call_lock = RW_LOCK_UNLOCKED;
> + 
> + long asmlinkage sys_putpmsg(int fd, void *ctlptr, void *datptr, int band, int flags)
> + {
> + 	int ret = -ENOSYS;
> + 	read_lock(&streams_call_lock);

Really you cannot use a spinlock here, because that would disallow
do_putpmsg from ever sleeping. Please review the mails I wrote 
earlier.

Either use atomic_inc(&some_counter) or a rw semaphore. I would likely
choose the atomic_inc. Alternatively you could use the RCU infrastructure
and stick a synchronize_kernel into module unload, then everything
would be fine too.

> + 	if (do_putpmsg)
> + 		ret = (*do_putpmsg) (fd, ctrlptr, datptr, band, flags);
> + 	read_unlock(&streams_call_lock);
> + 	return ret;

Same problem in the other stubs.

-Andi