Add MAP_NOSYNC feature to mmap(), and MADV_NOSYNC and MADV_AUTOSYNC to

    madvise().

    This feature prevents the update daemon from gratuitously flushing
    dirty pages associated with a mapped file-backed region of memory.  The
    system pager will still page the memory as necessary and the VM system
    will still be fully coherent with the filesystem.  Modifications made
    by other means to the same area of memory, for example by write(), are
    unaffected.  The feature works on a page-granularity basis.

    MAP_NOSYNC allows one to use mmap() to share memory between processes
    without incuring any significant filesystem overhead, putting it in
    the same performance category as SysV Shared memory and anonymous memory.

Reviewed by: julian, alc, dg

1 files changed, 31 insertions, 4 deletions

diff --git a/sys/vm/vm_object.c b/sys/vm/vm_object.c
index 1df9faf..6da6c9b 100644
--- a/sys/vm/vm_object.c
+++ b/sys/vm/vm_object.c
@@ -478,8 +478,10 @@ vm_object_terminate(object)
 /*
  *	vm_object_page_clean
  *
- *	Clean all dirty pages in the specified range of object.
- *	Leaves page on whatever queue it is currently on.
+ *	Clean all dirty pages in the specified range of object.  Leaves page 
+ * 	on whatever queue it is currently on.   If NOSYNC is set then do not
+ *	write out pages with PG_NOSYNC set (originally comes from MAP_NOSYNC),
+ *	leaving the object dirty.
  *
  *	Odd semantics: if start == end, we clean everything.
  *
@@ -503,6 +505,7 @@ vm_object_page_clean(object, start, end, flags)
 	int chkb;
 	int maxb;
 	int i;
+	int clearobjflags;
 	int pagerflags;
 	vm_page_t maf[vm_pageout_page_count];
 	vm_page_t mab[vm_pageout_page_count];
@@ -527,12 +530,26 @@ vm_object_page_clean(object, start, end, flags)
 		tend = end;
 	}
 
+	/*
+	 * Generally set CLEANCHK interlock and make the page read-only so
+	 * we can then clear the object flags.
+	 *
+	 * However, if this is a nosync mmap then the object is likely to 
+	 * stay dirty so do not mess with the page and do not clear the
+	 * object flags.
+	 */
+
+	clearobjflags = 1;
+
 	for(p = TAILQ_FIRST(&object->memq); p; p = TAILQ_NEXT(p, listq)) {
 		vm_page_flag_set(p, PG_CLEANCHK);
-		vm_page_protect(p, VM_PROT_READ);
+		if ((flags & OBJPC_NOSYNC) && (p->flags & PG_NOSYNC))
+			clearobjflags = 0;
+		else
+			vm_page_protect(p, VM_PROT_READ);
 	}
 
-	if ((tstart == 0) && (tend == object->size)) {
+	if (clearobjflags && (tstart == 0) && (tend == object->size)) {
 		vm_object_clear_flag(object, OBJ_WRITEABLE|OBJ_MIGHTBEDIRTY);
 	}
 
@@ -557,6 +574,16 @@ rescan:
 			continue;
 		}
 
+		/*
+		 * If we have been asked to skip nosync pages and this is a
+		 * nosync page, skip it.  Note that the object flags were
+		 * not cleared in this case so we do not have to set them.
+		 */
+		if ((flags & OBJPC_NOSYNC) && (p->flags & PG_NOSYNC)) {
+			vm_page_flag_clear(p, PG_CLEANCHK);
+			continue;
+		}
+
 		s = splvm();
 		while (vm_page_sleep_busy(p, TRUE, "vpcwai")) {
 			if (object->generation != curgeneration) {