Fixing gnumachs vm_map

[Justus Winter]

Hi :)

there is a bug affecting vm_map.  Contrary to the documentation,
address is not ignored if anywhere is given, leading to spurious
KERN_NO_SPACE errors.  Test case below.

I'm not sure the fix is correct, maybe we have to do

if (... || start + size > map->max_offset)...

or something.  Please have a look.

I have one Hurd patch that consistently triggers this while copying
files, and one that sometimes triggers this while compiling random
stuff.  I cannot believe this isn't causing more trouble all over the
place...

Justus

/* The documentation of vm_map says:

   kern_return_t vm_map (vm_task_t target_task,
                         vm_address_t *address,
                         ..., boolean_t anywhere, ...)

   [...] target_task is the task to be affected. The starting address
   is address. If the anywhere option is used, this address is
   ignored. [...]

   But:

   % ./gnumach_vm_map_bug
   *addr is (nil)
   *addr is 0xffffffff
   gnumach_vm_map_bug: gnumach_vm_map_bug.c:23: map: Unexpected error: 
(os/kern) no space available.

*/

#define _GNU_SOURCE
#include <assert.h>
#include <errno.h>
#include <mach.h>
#include <stdio.h>

void
map (vm_address_t *addr)
{
  error_t err;
  printf ("*addr is %p\n", *addr);
  err = vm_map (mach_task_self (),
                addr,
                1024,
                0, /* mask */
                1, /* anywhere */
                MACH_PORT_NULL,
                0, /* offset */
                0, /* copy */
                VM_PROT_READ,
                VM_PROT_READ,
                VM_INHERIT_NONE);
  assert_perror (err);
}

int
main (int argv, char **argc)
{
  vm_address_t addr;
  addr = 0;
  map (&addr);
  addr = ~0;
  map (&addr);
  return 0;
}