vm_map.c

open bsd vm device design

		vm_map_entry_delete(map, entry);
		entry = next;
	}
	return(KERN_SUCCESS);
}

/*
 *	vm_map_remove:
 *
 *	Remove the given address range from the target map.
 *	This is the exported form of vm_map_delete.
 */
int
vm_map_remove(map, start, end)
	register vm_map_t	map;
	register vm_offset_t	start;
	register vm_offset_t	end;
{
	register int		result;

	vm_map_lock(map);
	VM_MAP_RANGE_CHECK(map, start, end);
	result = vm_map_delete(map, start, end);
	vm_map_unlock(map);

	return(result);
}

/*
 *	vm_map_check_protection:
 *
 *	Assert that the target map allows the specified
 *	privilege on the entire address region given.
 *	The entire region must be allocated.
 */
boolean_t
vm_map_check_protection(map, start, end, protection)
	register vm_map_t	map;
	register vm_offset_t	start;
	register vm_offset_t	end;
	register vm_prot_t	protection;
{
	register vm_map_entry_t	entry;
	vm_map_entry_t		tmp_entry;

	if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
		return(FALSE);
	}

	entry = tmp_entry;

	while (start < end) {
		if (entry == &map->header) {
			return(FALSE);
		}

		/*
		 *	No holes allowed!
		 */

		if (start < entry->start) {
			return(FALSE);
		}

		/*
		 * Check protection associated with entry.
		 */

		if ((entry->protection & protection) != protection) {
			return(FALSE);
		}

		/* go to next entry */

		start = entry->end;
		entry = entry->next;
	}
	return(TRUE);
}

/*
 *	vm_map_copy_entry:
 *
 *	Copies the contents of the source entry to the destination
 *	entry.  The entries *must* be aligned properly.
 */
void
vm_map_copy_entry(src_map, dst_map, src_entry, dst_entry)
	vm_map_t		src_map, dst_map;
	register vm_map_entry_t	src_entry, dst_entry;
{
	vm_object_t	temp_object;

	if (src_entry->is_sub_map || dst_entry->is_sub_map)
		return;

	if (dst_entry->object.vm_object != NULL &&
	    (dst_entry->object.vm_object->flags & OBJ_INTERNAL) == 0)
		printf("vm_map_copy_entry: copying over permanent data!\n");

	/*
	 *	If our destination map was wired down,
	 *	unwire it now.
	 */

	if (dst_entry->wired_count != 0)
		vm_map_entry_unwire(dst_map, dst_entry);

	/*
	 *	If we're dealing with a sharing map, we
	 *	must remove the destination pages from
	 *	all maps (since we cannot know which maps
	 *	this sharing map belongs in).
	 */

	if (dst_map->is_main_map)
		pmap_remove(dst_map->pmap, dst_entry->start, dst_entry->end);
	else
		vm_object_pmap_remove(dst_entry->object.vm_object,
			dst_entry->offset,
			dst_entry->offset +
				(dst_entry->end - dst_entry->start));

	if (src_entry->wired_count == 0) {

		boolean_t	src_needs_copy;

		/*
		 *	If the source entry is marked needs_copy,
		 *	it is already write-protected.
		 */
		if (!src_entry->needs_copy) {

			boolean_t	su;

			/*
			 *	If the source entry has only one mapping,
			 *	we can just protect the virtual address
			 *	range.
			 */
			if (!(su = src_map->is_main_map)) {
				simple_lock(&src_map->ref_lock);
				su = (src_map->ref_count == 1);
				simple_unlock(&src_map->ref_lock);
			}

			if (su) {
				pmap_protect(src_map->pmap,
					src_entry->start,
					src_entry->end,
					src_entry->protection & ~VM_PROT_WRITE);
			}
			else {
				vm_object_pmap_copy(src_entry->object.vm_object,
					src_entry->offset,
					src_entry->offset + (src_entry->end
							    -src_entry->start));
			}
		}

		/*
		 *	Make a copy of the object.
		 */
		temp_object = dst_entry->object.vm_object;
		vm_object_copy(src_entry->object.vm_object,
				src_entry->offset,
				(vm_size_t)(src_entry->end -
					    src_entry->start),
				&dst_entry->object.vm_object,
				&dst_entry->offset,
				&src_needs_copy);
		/*
		 *	If we didn't get a copy-object now, mark the
		 *	source map entry so that a shadow will be created
		 *	to hold its changed pages.
		 */
		if (src_needs_copy)
			src_entry->needs_copy = TRUE;

		/*
		 *	The destination always needs to have a shadow
		 *	created.
		 */
		dst_entry->needs_copy = TRUE;

		/*
		 *	Mark the entries copy-on-write, so that write-enabling
		 *	the entry won't make copy-on-write pages writable.
		 */
		src_entry->copy_on_write = TRUE;
		dst_entry->copy_on_write = TRUE;
		/*
		 *	Get rid of the old object.
		 */
		vm_object_deallocate(temp_object);

		pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
			dst_entry->end - dst_entry->start, src_entry->start);
	}
	else {
		/*
		 *	Of course, wired down pages can't be set copy-on-write.
		 *	Cause wired pages to be copied into the new
		 *	map by simulating faults (the new pages are
		 *	pageable)
		 */
		vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry);
	}
}

/*
 *	vm_map_copy:
 *
 *	Perform a virtual memory copy from the source
 *	address map/range to the destination map/range.
 *
 *	If src_destroy or dst_alloc is requested,
 *	the source and destination regions should be
 *	disjoint, not only in the top-level map, but
 *	in the sharing maps as well.  [The best way
 *	to guarantee this is to use a new intermediate
 *	map to make copies.  This also reduces map
 *	fragmentation.]
 */
int
vm_map_copy(dst_map, src_map,
			  dst_addr, len, src_addr,
			  dst_alloc, src_destroy)
	vm_map_t	dst_map;
	vm_map_t	src_map;
	vm_offset_t	dst_addr;
	vm_size_t	len;
	vm_offset_t	src_addr;
	boolean_t	dst_alloc;
	boolean_t	src_destroy;
{
	register
	vm_map_entry_t	src_entry;
	register
	vm_map_entry_t	dst_entry;
	vm_map_entry_t	tmp_entry;
	vm_offset_t	src_start;
	vm_offset_t	src_end;
	vm_offset_t	dst_start;
	vm_offset_t	dst_end;
	vm_offset_t	src_clip;
	vm_offset_t	dst_clip;
	int		result;
	boolean_t	old_src_destroy;

	/*
	 *	XXX While we figure out why src_destroy screws up,
	 *	we'll do it by explicitly vm_map_delete'ing at the end.
	 */

	old_src_destroy = src_destroy;
	src_destroy = FALSE;

	/*
	 *	Compute start and end of region in both maps
	 */

	src_start = src_addr;
	src_end = src_start + len;
	dst_start = dst_addr;
	dst_end = dst_start + len;

	/*
	 *	Check that the region can exist in both source
	 *	and destination.
	 */

	if ((dst_end < dst_start) || (src_end < src_start))
		return(KERN_NO_SPACE);

	/*
	 *	Lock the maps in question -- we avoid deadlock
	 *	by ordering lock acquisition by map value
	 */

	if (src_map == dst_map) {
		vm_map_lock(src_map);
	}
	else if ((long) src_map < (long) dst_map) {
	 	vm_map_lock(src_map);
		vm_map_lock(dst_map);
	} else {
		vm_map_lock(dst_map);
	 	vm_map_lock(src_map);
	}

	result = KERN_SUCCESS;

	/*
	 *	Check protections... source must be completely readable and
	 *	destination must be completely writable.  [Note that if we're
	 *	allocating the destination region, we don't have to worry
	 *	about protection, but instead about whether the region
	 *	exists.]
	 */

	if (src_map->is_main_map && dst_map->is_main_map) {
		if (!vm_map_check_protection(src_map, src_start, src_end,
					VM_PROT_READ)) {
			result = KERN_PROTECTION_FAILURE;
			goto Return;
		}

		if (dst_alloc) {
			/* XXX Consider making this a vm_map_find instead */
			if ((result = vm_map_insert(dst_map, NULL,
					(vm_offset_t) 0, dst_start, dst_end)) != KERN_SUCCESS)
				goto Return;
		}
		else if (!vm_map_check_protection(dst_map, dst_start, dst_end,
					VM_PROT_WRITE)) {
			result = KERN_PROTECTION_FAILURE;
			goto Return;
		}
	}

	/*
	 *	Find the start entries and clip.
	 *
	 *	Note that checking protection asserts that the
	 *	lookup cannot fail.
	 *
	 *	Also note that we wait to do the second lookup
	 *	until we have done the first clip, as the clip
	 *	may affect which entry we get!
	 */

	(void) vm_map_lookup_entry(src_map, src_addr, &tmp_entry);
	src_entry = tmp_entry;
	vm_map_clip_start(src_map, src_entry, src_start);

	(void) vm_map_lookup_entry(dst_map, dst_addr, &tmp_entry);
	dst_entry = tmp_entry;
	vm_map_clip_start(dst_map, dst_entry, dst_start);

	/*
	 *	If both source and destination entries are the same,
	 *	retry the first lookup, as it may have changed.
	 */

	if (src_entry == dst_entry) {
		(void) vm_map_lookup_entry(src_map, src_addr, &tmp_entry);
		src_entry = tmp_entry;
	}

	/*
	 *	If source and destination entries are still the same,
	 *	a null copy is being performed.
	 */

	if (src_entry == dst_entry)
		goto Return;

	/*
	 *	Go through entries until we get to the end of the
	 *	region.
	 */

	while (src_start < src_end) {
		/*
		 *	Clip the entries to the endpoint of the entire region.
		 */

		vm_map_clip_end(src_map, src_entry, src_end);
		vm_map_clip_end(dst_map, dst_entry, dst_end);

		/*
		 *	Clip each entry to the endpoint of the other entry.
		 */

		src_clip = src_entry->start + (dst_entry->end - dst_entry->start);
		vm_map_clip_end(src_map, src_entry, src_clip);

		dst_clip = dst_entry->start + (src_entry->end - src_entry->start);
		vm_map_clip_end(dst_map, dst_entry, dst_clip);

		/*
		 *	Both entries now match in size and relative endpoints.
		 *
		 *	If both entries refer to a VM object, we can
		 *	deal with them now.
		 */

		if (!src_entry->is_a_map && !dst_entry->is_a_map) {
			vm_map_copy_entry(src_map, dst_map, src_entry,
						dst_entry);
		}
		else {
			register vm_map_t	new_dst_map;
			vm_offset_t		new_dst_start;
			vm_size_t		new_size;
			vm_map_t		new_src_map;
			vm_offset_t		new_src_start;

			/*
			 *	We have to follow at least one sharing map.
			 */

			new_size = (dst_entry->end - dst_entry->start);

			if (src_entry->is_a_map) {
				new_src_map = src_entry->object.share_map;
				new_src_start = src_entry->offset;
			}
			else {
			 	new_src_map = src_map;
				new_src_start = src_entry->start;
				vm_map_set_recursive(&src_map->lock);
			}

			if (dst_entry->is_a_map) {
			    	vm_offset_t	new_dst_end;

				new_dst_map = dst_entry->object.share_map;
				new_dst_start = dst_entry->offset;

				/*
				 *	Since the destination sharing entries
				 *	will be merely deallocated, we can
				 *	do that now, and replace the region
				 *	with a null object.  [This prevents
				 *	splitting the source map to match
				 *	the form of the destination map.]
				 *	Note that we can only do so if the
				 *	source and destination do not overlap.
				 */

				new_dst_end = new_dst_start + new_size;

				if (new_dst_map != new_src_map) {
					vm_map_lock(new_dst_map);
					(void) vm_map_delete(new_dst_map,
							new_dst_start,
							new_dst_end);
					(void) vm_map_insert(new_dst_map,
							NULL,
							(vm_offset_t) 0,
							new_dst_start,
							new_dst_end);
					vm_map_unlock(new_dst_map);
				}
			}
			else {
			 	new_dst_map = dst_map;
				new_dst_start = dst_entry->start;
				vm_map_set_recursive(&dst_map->lock);
			}

			/*
			 *	Recursively copy the sharing map.
			 */

			(void) vm_map_copy(new_dst_map, new_src_map,
				new_dst_start, new_size, new_src_start,
				FALSE, FALSE);

			if (dst_map == new_dst_map)
				vm_map_clear_recursive(&dst_map->lock);
			if (src_map == new_src_map)
				vm_map_clear_recursive(&src_map->lock);
		}

		/*
		 *	Update variables for next pass through the loop.
		 */

		src_start = src_entry->end;
		src_entry = src_entry->next;
		dst_start = dst_entry->end;
		dst_entry = dst_entry->next;

		/*
		 *	If the source is to be destroyed, here is the
		 *	place to do it.
		 */

		if (src_destroy && src_map->is_main_map &&
						dst_map->is_main_map)
			vm_map_entry_delete(src_map, src_entry->prev);
	}

	/*
	 *	Update the physical maps as appropriate
	 */

	if (src_map->is_main_map && dst_map->is_main_map) {
		if (src_destroy)
			pmap_remove(src_map->pmap, src_addr, src_addr + len);
	}

	/*
	 *	Unlock the maps
	 */

	Return: ;

	if (old_src_destroy)
		vm_map_delete(src_map, src_addr, src_addr + len);

	vm_map_unlock(src_map);
	if (src_map != dst_map)
		vm_map_unlock(dst_map);

	return(result);
}

/*
 * vmspace_fork:
 * Create a new process vmspace structure and vm_map
 * based on those of an existing process.  The new map
 * is based on the old map, according to the inheritance
 * values on the regions in that map.
 *
 * The source map must not be locked.
 */
struct vmspace *
vmspace_fork(vm1)
	register struct vmspace *vm1;
{
	register struct vmspace *vm2;
	vm_map_t	old_map = &vm1->vm_map;
	vm_map_t	new_map;
	vm_map_entry_t	old_entry;
	vm_map_entry_t	new_entry;
	pmap_t		new_pmap;