sys_modify_ldt extension (default_ldt)

Hi,

I am trying to integrate binfmt_xout.c into kernel 2.4 as part of the linux-abi project (formerly known as iBCS). For old Xenix 286 binaries the lcall7 gate needs to part of the LDT.

In kernels 2.0 sys_modify_ldt(0,...) used to return the default_ldt (with lcall7 gate) if there were no segments set up. This behaviour changed in kernels 2.2 . As a result of a discussion with Linus, David Bruce wrote a patch for binfmt_xout.c tweaking with gdt and current->tss.ldt to get the address of default_ldt. This patch does not work any more with kernels 2.4 as tss vanished from task_struct.

I do see 4 ways to cope with this problem:

a) extend sys_modify_ldt with a function to retrieve the default_ldt. I did this for testing (see attached diff for arch/i386/kernel/ldt.c ).

b) do some work an Davids patch but this is kind of magic for me :-) (see attached default_ldt patch)

c) loose the option to compile binfmt_xout (and the rest of linux-abi) as module and simply use the symbol default_ldt. I dint't try that.

d) Forget about those old fashioned 286 binaries. This option will make some linux users feel sad, as they run these progs for their daily business.

Joerg

------------------------------------------------------------------------------
Joerg Ahrens                                              _/           
Koenigsberger Strasse 32                                _/_/
31226 Peine                                           _/  _/
Tel.: 05171/57308                             _/    _/_/_/_/
e-mail: [email protected]                _/_/_/_/      _/
------------------------------------------------------------------------------

@@ -44,7 +44,24 @@
 out:
 	return err;
 }
+static int read_default_ldt(void * ptr, unsigned long bytecount)
+{
+	int err;
+	unsigned long size;
+	void *address;
+
+	err = 0;
+	address = &default_ldt[0];
+	size = sizeof(struct desc_struct);
+	if (size > bytecount)
+		size = bytecount;
+
+	err = size;
+	if (copy_to_user(ptr, address, size))
+		err = -EFAULT;
 
+	return err;
+}
 static int write_ldt(void * ptr, unsigned long bytecount, int oldmode)
 {
 	struct mm_struct * mm = current->mm;

@@ -156,6 +173,9 @@
 		break;
 	case 1:
 		ret = write_ldt(ptr, bytecount, 1);
+		break;
+	case 2:
+		ret = read_default_ldt(ptr, bytecount);
 		break;
 	case 0x11:
 		ret = write_ldt(ptr, bytecount, 0);

struct desc_struct def_ldt;
unsigned long *lp, *lp2;

asm volatile ( "sgdt __gdt+2" );

lp = (unsigned long *)(__gdt[1] + current->tss.ldt );

lp2 = (unsigned long *)(((*lp >> 16) & 0x0000ffff)

           | (*(lp+1) & 0xff000000)
           | ((*(lp+1) << 16) & 0x00ff0000));

def_ldt.a = *lp2;
def_ldt.b = *(lp2+1);

Hi Joerg,

Looks sane to me.

As the linux-abi maintainer I do not think that's a good idea..

If possible at all I'll vote for having 286 support.

I'll integrate the patch into the linux-abi tree if you also send me the other changes (mostly binfmt_xout.c changes I suppose).

Christoph

Of course it doesn't work. We've performed a software upgrade.