vm_map.c

open bsd vm device design

			else
			 	pmap_protect(map->pmap, current->start,
					current->end,
					current->protection & MASK(entry));
#undef	max
#undef	MASK
		}
		current = current->next;
	}

	vm_map_unlock(map);
	return(KERN_SUCCESS);
}

/*
 *	vm_map_inherit:
 *
 *	Sets the inheritance of the specified address
 *	range in the target map.  Inheritance
 *	affects how the map will be shared with
 *	child maps at the time of vm_map_fork.
 */
int
vm_map_inherit(map, start, end, new_inheritance)
	register vm_map_t	map;
	register vm_offset_t	start;
	register vm_offset_t	end;
	register vm_inherit_t	new_inheritance;
{
	register vm_map_entry_t	entry;
	vm_map_entry_t	temp_entry;

	switch (new_inheritance) {
	case VM_INHERIT_NONE:
	case VM_INHERIT_COPY:
	case VM_INHERIT_SHARE:
		break;
	default:
		return(KERN_INVALID_ARGUMENT);
	}

	vm_map_lock(map);

	VM_MAP_RANGE_CHECK(map, start, end);

	if (vm_map_lookup_entry(map, start, &temp_entry)) {
		entry = temp_entry;
		vm_map_clip_start(map, entry, start);
	}
	else
		entry = temp_entry->next;

	while ((entry != &map->header) && (entry->start < end)) {
		vm_map_clip_end(map, entry, end);

		entry->inheritance = new_inheritance;

		entry = entry->next;
	}

	vm_map_unlock(map);
	return(KERN_SUCCESS);
}

/*
 *	vm_map_pageable:
 *
 *	Sets the pageability of the specified address
 *	range in the target map.  Regions specified
 *	as not pageable require locked-down physical
 *	memory and physical page maps.
 *
 *	The map must not be locked, but a reference
 *	must remain to the map throughout the call.
 */
int
vm_map_pageable(map, start, end, new_pageable)
	register vm_map_t	map;
	register vm_offset_t	start;
	register vm_offset_t	end;
	register boolean_t	new_pageable;
{
	register vm_map_entry_t	entry;
	vm_map_entry_t		start_entry;
	register vm_offset_t	failed;
	int			rv;

	vm_map_lock(map);

	VM_MAP_RANGE_CHECK(map, start, end);

	/*
	 *	Only one pageability change may take place at one
	 *	time, since vm_fault assumes it will be called
	 *	only once for each wiring/unwiring.  Therefore, we
	 *	have to make sure we're actually changing the pageability
	 *	for the entire region.  We do so before making any changes.
	 */

	if (vm_map_lookup_entry(map, start, &start_entry) == FALSE) {
		vm_map_unlock(map);
		return(KERN_INVALID_ADDRESS);
	}
	entry = start_entry;

	/*
	 *	Actions are rather different for wiring and unwiring,
	 *	so we have two separate cases.
	 */

	if (new_pageable) {

		vm_map_clip_start(map, entry, start);

		/*
		 *	Unwiring.  First ensure that the range to be
		 *	unwired is really wired down and that there
		 *	are no holes.
		 */
		while ((entry != &map->header) && (entry->start < end)) {

		    if (entry->wired_count == 0 ||
			(entry->end < end &&
			 (entry->next == &map->header ||
			  entry->next->start > entry->end))) {
			vm_map_unlock(map);
			return(KERN_INVALID_ARGUMENT);
		    }
		    entry = entry->next;
		}

		/*
		 *	Now decrement the wiring count for each region.
		 *	If a region becomes completely unwired,
		 *	unwire its physical pages and mappings.
		 */
		vm_map_set_recursive(&map->lock);

		entry = start_entry;
		while ((entry != &map->header) && (entry->start < end)) {
		    vm_map_clip_end(map, entry, end);

		    entry->wired_count--;
		    if (entry->wired_count == 0)
			vm_fault_unwire(map, entry->start, entry->end);

		    entry = entry->next;
		}
		vm_map_clear_recursive(&map->lock);
	}

	else {
		/*
		 *	Wiring.  We must do this in two passes:
		 *
		 *	1.  Holding the write lock, we create any shadow
		 *	    or zero-fill objects that need to be created.
		 *	    Then we clip each map entry to the region to be
		 *	    wired and increment its wiring count.  We
		 *	    create objects before clipping the map entries
		 *	    to avoid object proliferation.
		 *
		 *	2.  We downgrade to a read lock, and call
		 *	    vm_fault_wire to fault in the pages for any
		 *	    newly wired area (wired_count is 1).
		 *
		 *	Downgrading to a read lock for vm_fault_wire avoids
		 *	a possible deadlock with another thread that may have
		 *	faulted on one of the pages to be wired (it would mark
		 *	the page busy, blocking us, then in turn block on the
		 *	map lock that we hold).  Because of problems in the
		 *	recursive lock package, we cannot upgrade to a write
		 *	lock in vm_map_lookup.  Thus, any actions that require
		 *	the write lock must be done beforehand.  Because we
		 *	keep the read lock on the map, the copy-on-write status
		 *	of the entries we modify here cannot change.
		 */

		/*
		 *	Pass 1.
		 */
		while ((entry != &map->header) && (entry->start < end)) {
		    if (entry->wired_count == 0) {

			/*
			 *	Perform actions of vm_map_lookup that need
			 *	the write lock on the map: create a shadow
			 *	object for a copy-on-write region, or an
			 *	object for a zero-fill region.
			 *
			 *	We don't have to do this for entries that
			 *	point to sharing maps, because we won't hold
			 *	the lock on the sharing map.
			 */
			if (!entry->is_a_map) {
			    if (entry->needs_copy &&
				((entry->protection & VM_PROT_WRITE) != 0)) {

				vm_object_shadow(&entry->object.vm_object,
						&entry->offset,
						(vm_size_t)(entry->end
							- entry->start));
				entry->needs_copy = FALSE;
			    }
			    else if (entry->object.vm_object == NULL) {
				entry->object.vm_object =
				    vm_object_allocate((vm_size_t)(entry->end
				    			- entry->start));
				entry->offset = (vm_offset_t)0;
			    }
			}
		    }
		    vm_map_clip_start(map, entry, start);
		    vm_map_clip_end(map, entry, end);
		    entry->wired_count++;

		    /*
		     * Check for holes
		     */
		    if (entry->end < end &&
			(entry->next == &map->header ||
			 entry->next->start > entry->end)) {
			/*
			 *	Found one.  Object creation actions
			 *	do not need to be undone, but the
			 *	wired counts need to be restored.
			 */
			while (entry != &map->header && entry->end > start) {
			    entry->wired_count--;
			    entry = entry->prev;
			}
			vm_map_unlock(map);
			return(KERN_INVALID_ARGUMENT);
		    }
		    entry = entry->next;
		}

		/*
		 *	Pass 2.
		 */

		/*
		 * HACK HACK HACK HACK
		 *
		 * If we are wiring in the kernel map or a submap of it,
		 * unlock the map to avoid deadlocks.  We trust that the
		 * kernel threads are well-behaved, and therefore will
		 * not do anything destructive to this region of the map
		 * while we have it unlocked.  We cannot trust user threads
		 * to do the same.
		 *
		 * HACK HACK HACK HACK
		 */
		if (vm_map_pmap(map) == kernel_pmap) {
		    vm_map_unlock(map);		/* trust me ... */
		}
		else {
		    vm_map_set_recursive(&map->lock);
		    lockmgr(&map->lock, LK_DOWNGRADE, (void *)0, curproc);
		}

		rv = 0;
		entry = start_entry;
		while (entry != &map->header && entry->start < end) {
		    /*
		     * If vm_fault_wire fails for any page we need to
		     * undo what has been done.  We decrement the wiring
		     * count for those pages which have not yet been
		     * wired (now) and unwire those that have (later).
		     *
		     * XXX this violates the locking protocol on the map,
		     * needs to be fixed.
		     */
		    if (rv)
			entry->wired_count--;
		    else if (entry->wired_count == 1) {
			rv = vm_fault_wire(map, entry->start, entry->end);
			if (rv) {
			    failed = entry->start;
			    entry->wired_count--;
			}
		    }
		    entry = entry->next;
		}

		if (vm_map_pmap(map) == kernel_pmap) {
		    vm_map_lock(map);
		}
		else {
		    vm_map_clear_recursive(&map->lock);
		}
		if (rv) {
		    vm_map_unlock(map);
		    (void) vm_map_pageable(map, start, failed, TRUE);
		    return(rv);
		}
	}

	vm_map_unlock(map);

	return(KERN_SUCCESS);
}

/*
 * vm_map_clean
 *
 * Push any dirty cached pages in the address range to their pager.
 * If syncio is TRUE, dirty pages are written synchronously.
 * If invalidate is TRUE, any cached pages are freed as well.
 *
 * Returns an error if any part of the specified range is not mapped.
 */
int
vm_map_clean(map, start, end, syncio, invalidate)
	vm_map_t	map;
	vm_offset_t	start;
	vm_offset_t	end;
	boolean_t	syncio;
	boolean_t	invalidate;
{
	register vm_map_entry_t current;
	vm_map_entry_t entry;
	vm_size_t size;
	vm_object_t object;
	vm_offset_t offset;

	vm_map_lock_read(map);
	VM_MAP_RANGE_CHECK(map, start, end);
	if (!vm_map_lookup_entry(map, start, &entry)) {
		vm_map_unlock_read(map);
		return(KERN_INVALID_ADDRESS);
	}

	/*
	 * Make a first pass to check for holes.
	 */
	for (current = entry; current->start < end; current = current->next) {
		if (current->is_sub_map) {
			vm_map_unlock_read(map);
			return(KERN_INVALID_ARGUMENT);
		}
		if (end > current->end &&
		    (current->next == &map->header ||
		     current->end != current->next->start)) {
			vm_map_unlock_read(map);
			return(KERN_INVALID_ADDRESS);
		}
	}

	/*
	 * Make a second pass, cleaning/uncaching pages from the indicated
	 * objects as we go.
	 */
	for (current = entry; current->start < end; current = current->next) {
		offset = current->offset + (start - current->start);
		size = (end <= current->end ? end : current->end) - start;
		if (current->is_a_map) {
			register vm_map_t smap;
			vm_map_entry_t tentry;
			vm_size_t tsize;

			smap = current->object.share_map;
			vm_map_lock_read(smap);
			(void) vm_map_lookup_entry(smap, offset, &tentry);
			tsize = tentry->end - offset;
			if (tsize < size)
				size = tsize;
			object = tentry->object.vm_object;
			offset = tentry->offset + (offset - tentry->start);
			vm_object_lock(object);
			vm_map_unlock_read(smap);
		} else {
			object = current->object.vm_object;
			vm_object_lock(object);
		}
		/*
		 * Flush pages if writing is allowed.
		 * XXX should we continue on an error?
		 */
		if ((current->protection & VM_PROT_WRITE) &&
		    !vm_object_page_clean(object, offset, offset+size,
					  syncio, FALSE)) {
			vm_object_unlock(object);
			vm_map_unlock_read(map);
			return(KERN_FAILURE);
		}
		if (invalidate)
			vm_object_page_remove(object, offset, offset+size);
		vm_object_unlock(object);
		start += size;
	}

	vm_map_unlock_read(map);
	return(KERN_SUCCESS);
}

/*
 *	vm_map_entry_unwire:	[ internal use only ]
 *
 *	Make the region specified by this entry pageable.
 *
 *	The map in question should be locked.
 *	[This is the reason for this routine's existence.]
 */
void
vm_map_entry_unwire(map, entry)
	vm_map_t		map;
	register vm_map_entry_t	entry;
{
	vm_fault_unwire(map, entry->start, entry->end);
	entry->wired_count = 0;
}

/*
 *	vm_map_entry_delete:	[ internal use only ]
 *
 *	Deallocate the given entry from the target map.
 */		
void
vm_map_entry_delete(map, entry)
	register vm_map_t	map;
	register vm_map_entry_t	entry;
{
	if (entry->wired_count != 0)
		vm_map_entry_unwire(map, entry);
		
	vm_map_entry_unlink(map, entry);
	map->size -= entry->end - entry->start;

	if (entry->is_a_map || entry->is_sub_map)
		vm_map_deallocate(entry->object.share_map);
	else
	 	vm_object_deallocate(entry->object.vm_object);

	vm_map_entry_dispose(map, entry);
}

/*
 *	vm_map_delete:	[ internal use only ]
 *
 *	Deallocates the given address range from the target
 *	map.
 *
 *	When called with a sharing map, removes pages from
 *	that region from all physical maps.
 */
int
vm_map_delete(map, start, end)
	register vm_map_t	map;
	vm_offset_t		start;
	register vm_offset_t	end;
{
	register vm_map_entry_t	entry;
	vm_map_entry_t		first_entry;

	/*
	 *	Find the start of the region, and clip it
	 */

	if (!vm_map_lookup_entry(map, start, &first_entry))
		entry = first_entry->next;
	else {
		entry = first_entry;
		vm_map_clip_start(map, entry, start);

		/*
		 *	Fix the lookup hint now, rather than each
		 *	time though the loop.
		 */

		SAVE_HINT(map, entry->prev);
	}

	/*
	 *	Save the free space hint
	 */

	if (map->first_free->start >= start)
		map->first_free = entry->prev;

	/*
	 *	Step through all entries in this region
	 */

	while ((entry != &map->header) && (entry->start < end)) {
		vm_map_entry_t		next;
		register vm_offset_t	s, e;
		register vm_object_t	object;

		vm_map_clip_end(map, entry, end);

		next = entry->next;
		s = entry->start;
		e = entry->end;

		/*
		 *	Unwire before removing addresses from the pmap;
		 *	otherwise, unwiring will put the entries back in
		 *	the pmap.
		 */

		object = entry->object.vm_object;
		if (entry->wired_count != 0)
			vm_map_entry_unwire(map, entry);

		/*
		 *	If this is a sharing map, we must remove
		 *	*all* references to this data, since we can't
		 *	find all of the physical maps which are sharing
		 *	it.
		 */

		if (object == kernel_object || object == kmem_object)
			vm_object_page_remove(object, entry->offset,
					entry->offset + (e - s));
		else if (!map->is_main_map)
			vm_object_pmap_remove(object,
					 entry->offset,
					 entry->offset + (e - s));
		else
			pmap_remove(map->pmap, s, e);

		/*
		 *	Delete the entry (which may delete the object)
		 *	only after removing all pmap entries pointing
		 *	to its pages.  (Otherwise, its page frames may
		 *	be reallocated, and any modify bits will be
		 *	set in the wrong object!)
		 */