- Rename vm_map_clean() to vm_map_sync(). This better reflects the fact

   that msync(2) is its only caller.
 - Migrate the parts of the old vm_map_clean() that examined the internals
   of a vm object to a new function vm_object_sync() that is implemented in
   vm_object.c.  At the same, introduce the necessary vm object locking so
   that vm_map_sync() and vm_object_sync() can be called without Giant.

Reviewed by:	tegge

5 files changed, 78 insertions, 61 deletions

diff --git a/sys/vm/vm_map.c b/sys/vm/vm_map.c
index 1586cb3..40db4c1 100644
--- a/sys/vm/vm_map.c
+++ b/sys/vm/vm_map.c
@@ -1946,7 +1946,7 @@ done:
 }
 
 /*
- * vm_map_clean
+ * vm_map_sync
  *
  * Push any dirty cached pages in the address range to their pager.
  * If syncio is TRUE, dirty pages are written synchronously.
@@ -1955,7 +1955,7 @@ done:
  * Returns an error if any part of the specified range is not mapped.
  */
 int
-vm_map_clean(
+vm_map_sync(
 	vm_map_t map,
 	vm_offset_t start,
 	vm_offset_t end,
@@ -1968,8 +1968,6 @@ vm_map_clean(
 	vm_object_t object;
 	vm_ooffset_t offset;
 
-	GIANT_REQUIRED;
-
 	vm_map_lock_read(map);
 	VM_MAP_RANGE_CHECK(map, start, end);
 	if (!vm_map_lookup_entry(map, start, &entry)) {
@@ -1993,9 +1991,11 @@ vm_map_clean(
 	}
 
 	if (invalidate) {
+		mtx_lock(&Giant);
 		vm_page_lock_queues();
 		pmap_remove(map->pmap, start, end);
 		vm_page_unlock_queues();
+		mtx_unlock(&Giant);
 	}
 	/*
 	 * Make a second pass, cleaning/uncaching pages from the indicated
@@ -2021,61 +2021,7 @@ vm_map_clean(
 		} else {
 			object = current->object.vm_object;
 		}
-		/*
-		 * Note that there is absolutely no sense in writing out
-		 * anonymous objects, so we track down the vnode object
-		 * to write out.
-		 * We invalidate (remove) all pages from the address space
-		 * anyway, for semantic correctness.
-		 *
-		 * note: certain anonymous maps, such as MAP_NOSYNC maps,
-		 * may start out with a NULL object.
-		 */
-		while (object && object->backing_object) {
-			object = object->backing_object;
-			offset += object->backing_object_offset;
-			if (object->size < OFF_TO_IDX(offset + size))
-				size = IDX_TO_OFF(object->size) - offset;
-		}
-		if (object && (object->type == OBJT_VNODE) &&
-		    (current->protection & VM_PROT_WRITE)) {
-			/*
-			 * Flush pages if writing is allowed, invalidate them
-			 * if invalidation requested.  Pages undergoing I/O
-			 * will be ignored by vm_object_page_remove().
-			 *
-			 * We cannot lock the vnode and then wait for paging
-			 * to complete without deadlocking against vm_fault.
-			 * Instead we simply call vm_object_page_remove() and
-			 * allow it to block internally on a page-by-page
-			 * basis when it encounters pages undergoing async
-			 * I/O.
-			 */
-			int flags;
-
-			vm_object_reference(object);
-			vn_lock(object->handle, LK_EXCLUSIVE | LK_RETRY, curthread);
-			flags = (syncio || invalidate) ? OBJPC_SYNC : 0;
-			flags |= invalidate ? OBJPC_INVAL : 0;
-			VM_OBJECT_LOCK(object);
-			vm_object_page_clean(object,
-			    OFF_TO_IDX(offset),
-			    OFF_TO_IDX(offset + size + PAGE_MASK),
-			    flags);
-			VM_OBJECT_UNLOCK(object);
-			VOP_UNLOCK(object->handle, 0, curthread);
-			vm_object_deallocate(object);
-		}
-		if (object && invalidate &&
-		    ((object->type == OBJT_VNODE) ||
-		     (object->type == OBJT_DEVICE))) {
-			VM_OBJECT_LOCK(object);
-			vm_object_page_remove(object,
-			    OFF_TO_IDX(offset),
-			    OFF_TO_IDX(offset + size + PAGE_MASK),
-			    FALSE);
-			VM_OBJECT_UNLOCK(object);
-		}
+		vm_object_sync(object, offset, size, syncio, invalidate);
 		start += size;
 	}
 
diff --git a/sys/vm/vm_map.h b/sys/vm/vm_map.h
index 72e804d..241cba2 100644
--- a/sys/vm/vm_map.h
+++ b/sys/vm/vm_map.h
@@ -337,13 +337,13 @@ int vm_map_lookup (vm_map_t *, vm_offset_t, vm_prot_t, vm_map_entry_t *, vm_obje
     vm_pindex_t *, vm_prot_t *, boolean_t *);
 void vm_map_lookup_done (vm_map_t, vm_map_entry_t);
 boolean_t vm_map_lookup_entry (vm_map_t, vm_offset_t, vm_map_entry_t *);
-int vm_map_clean (vm_map_t, vm_offset_t, vm_offset_t, boolean_t, boolean_t);
 void vm_map_pmap_enter(vm_map_t map, vm_offset_t addr,
     vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags);
 int vm_map_protect (vm_map_t, vm_offset_t, vm_offset_t, vm_prot_t, boolean_t);
 int vm_map_remove (vm_map_t, vm_offset_t, vm_offset_t);
 void vm_map_startup (void);
 int vm_map_submap (vm_map_t, vm_offset_t, vm_offset_t, vm_map_t);
+int vm_map_sync(vm_map_t, vm_offset_t, vm_offset_t, boolean_t, boolean_t);
 int vm_map_madvise (vm_map_t, vm_offset_t, vm_offset_t, int);
 void vm_map_simplify_entry (vm_map_t, vm_map_entry_t);
 void vm_init2 (void);
diff --git a/sys/vm/vm_mmap.c b/sys/vm/vm_mmap.c
index a7dfc0d..158a443 100644
--- a/sys/vm/vm_mmap.c
+++ b/sys/vm/vm_mmap.c
@@ -574,7 +574,7 @@ msync(td, uap)
 	/*
 	 * Clean the pages and interpret the return value.
 	 */
-	rv = vm_map_clean(map, addr, addr + size, (flags & MS_ASYNC) == 0,
+	rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
 	    (flags & MS_INVALIDATE) != 0);
 
 done2:
diff --git a/sys/vm/vm_object.c b/sys/vm/vm_object.c
index abf4cf9..0099e52 100644
--- a/sys/vm/vm_object.c
+++ b/sys/vm/vm_object.c
@@ -964,6 +964,75 @@ vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int curgeneration,
 }
 
 /*
+ * Note that there is absolutely no sense in writing out
+ * anonymous objects, so we track down the vnode object
+ * to write out.
+ * We invalidate (remove) all pages from the address space
+ * for semantic correctness.
+ *
+ * Note: certain anonymous maps, such as MAP_NOSYNC maps,
+ * may start out with a NULL object.
+ */
+void
+vm_object_sync(vm_object_t object, vm_ooffset_t offset, vm_size_t size,
+    boolean_t syncio, boolean_t invalidate)
+{
+	vm_object_t backing_object;
+	struct vnode *vp;
+	int flags;
+
+	if (object == NULL)
+		return;
+	VM_OBJECT_LOCK(object);
+	while ((backing_object = object->backing_object) != NULL) {
+		VM_OBJECT_LOCK(backing_object);
+		VM_OBJECT_UNLOCK(object);
+		object = backing_object;
+		offset += object->backing_object_offset;
+		if (object->size < OFF_TO_IDX(offset + size))
+			size = IDX_TO_OFF(object->size) - offset;
+	}
+	/*
+	 * Flush pages if writing is allowed, invalidate them
+	 * if invalidation requested.  Pages undergoing I/O
+	 * will be ignored by vm_object_page_remove().
+	 *
+	 * We cannot lock the vnode and then wait for paging
+	 * to complete without deadlocking against vm_fault.
+	 * Instead we simply call vm_object_page_remove() and
+	 * allow it to block internally on a page-by-page
+	 * basis when it encounters pages undergoing async
+	 * I/O.
+	 */
+	if (object->type == OBJT_VNODE &&
+	    (object->flags & OBJ_MIGHTBEDIRTY) != 0) {
+		vp = object->handle;
+		VM_OBJECT_UNLOCK(object);
+		mtx_lock(&Giant);
+		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread);
+		flags = (syncio || invalidate) ? OBJPC_SYNC : 0;
+		flags |= invalidate ? OBJPC_INVAL : 0;
+		VM_OBJECT_LOCK(object);
+		vm_object_page_clean(object,
+		    OFF_TO_IDX(offset),
+		    OFF_TO_IDX(offset + size + PAGE_MASK),
+		    flags);
+		VM_OBJECT_UNLOCK(object);
+		VOP_UNLOCK(vp, 0, curthread);
+		mtx_unlock(&Giant);
+		VM_OBJECT_LOCK(object);
+	}
+	if ((object->type == OBJT_VNODE ||
+	     object->type == OBJT_DEVICE) && invalidate) {
+		vm_object_page_remove(object,
+		    OFF_TO_IDX(offset),
+		    OFF_TO_IDX(offset + size + PAGE_MASK),
+		    FALSE);
+	}
+	VM_OBJECT_UNLOCK(object);
+}
+
+/*
  *	vm_object_madvise:
  *
  *	Implements the madvise function at the object/page level.
diff --git a/sys/vm/vm_object.h b/sys/vm/vm_object.h
index 96d8666..bd4f44d 100644
--- a/sys/vm/vm_object.h
+++ b/sys/vm/vm_object.h
@@ -216,6 +216,8 @@ void vm_object_reference (vm_object_t);
 void vm_object_reference_locked(vm_object_t);
 void vm_object_shadow (vm_object_t *, vm_ooffset_t *, vm_size_t);
 void vm_object_split(vm_map_entry_t);
+void vm_object_sync(vm_object_t, vm_ooffset_t, vm_size_t, boolean_t,
+    boolean_t);
 void vm_object_madvise (vm_object_t, vm_pindex_t, int, int);
 #endif				/* _KERNEL */