3 files changed, 82 insertions, 11 deletions

diff --git a/sys/kern/vfs_bio.c b/sys/kern/vfs_bio.c
index 58e349147ba6..2979f61f24ce 100644
--- a/sys/kern/vfs_bio.c
+++ b/sys/kern/vfs_bio.c
@@ -98,7 +98,8 @@ static void vm_hold_free_pages(struct buf *bp, vm_offset_t from,
 		vm_offset_t to);
 static void vm_hold_load_pages(struct buf *bp, vm_offset_t from,
 		vm_offset_t to);
-static void vfs_page_set_valid(struct buf *bp, vm_ooffset_t off,
+static void vfs_page_set_valid(struct buf *bp, vm_ooffset_t off, vm_page_t m);
+static void vfs_page_set_validclean(struct buf *bp, vm_ooffset_t off,
 		vm_page_t m);
 static void vfs_clean_pages(struct buf *bp);
 static void vfs_setdirty(struct buf *bp);
@@ -3277,7 +3278,6 @@ bufdone_finish(struct buf *bp)
 		vm_object_t obj;
 		int iosize;
 		struct vnode *vp = bp->b_vp;
-		boolean_t are_queues_locked;
 
 		obj = bp->b_bufobj->bo_object;
 
@@ -3314,11 +3314,6 @@ bufdone_finish(struct buf *bp)
 		    !(bp->b_ioflags & BIO_ERROR)) {
 			bp->b_flags |= B_CACHE;
 		}
-		if (bp->b_iocmd == BIO_READ) {
-			vm_page_lock_queues();
-			are_queues_locked = TRUE;
-		} else
-			are_queues_locked = FALSE;
 		for (i = 0; i < bp->b_npages; i++) {
 			int bogusflag = 0;
 			int resid;
@@ -3354,6 +3349,9 @@ bufdone_finish(struct buf *bp)
 			 * only need to do this here in the read case.
 			 */
 			if ((bp->b_iocmd == BIO_READ) && !bogusflag && resid > 0) {
+				KASSERT((m->dirty & vm_page_bits(foff &
+				    PAGE_MASK, resid)) == 0, ("bufdone_finish:"
+				    " page %p has unexpected dirty bits", m));
 				vfs_page_set_valid(bp, foff, m);
 			}
 
@@ -3387,8 +3385,6 @@ bufdone_finish(struct buf *bp)
 			foff = (foff + PAGE_SIZE) & ~(off_t)PAGE_MASK;
 			iosize -= resid;
 		}
-		if (are_queues_locked)
-			vm_page_unlock_queues();
 		vm_object_pip_wakeupn(obj, 0);
 		VM_OBJECT_UNLOCK(obj);
 	}
@@ -3454,6 +3450,35 @@ vfs_unbusy_pages(struct buf *bp)
 static void
 vfs_page_set_valid(struct buf *bp, vm_ooffset_t off, vm_page_t m)
 {
+	vm_ooffset_t eoff;
+
+	/*
+	 * Compute the end offset, eoff, such that [off, eoff) does not span a
+	 * page boundary and eoff is not greater than the end of the buffer.
+	 * The end of the buffer, in this case, is our file EOF, not the
+	 * allocation size of the buffer.
+	 */
+	eoff = (off + PAGE_SIZE) & ~(vm_ooffset_t)PAGE_MASK;
+	if (eoff > bp->b_offset + bp->b_bcount)
+		eoff = bp->b_offset + bp->b_bcount;
+
+	/*
+	 * Set valid range.  This is typically the entire buffer and thus the
+	 * entire page.
+	 */
+	if (eoff > off)
+		vm_page_set_valid(m, off & PAGE_MASK, eoff - off);
+}
+
+/*
+ * vfs_page_set_validclean:
+ *
+ *	Set the valid bits and clear the dirty bits in a page based on the
+ *	supplied offset.   The range is restricted to the buffer's size.
+ */
+static void
+vfs_page_set_validclean(struct buf *bp, vm_ooffset_t off, vm_page_t m)
+{
 	vm_ooffset_t soff, eoff;
 
 	mtx_assert(&vm_page_queue_mtx, MA_OWNED);
@@ -3545,7 +3570,7 @@ retry:
 		 */
 		if (clear_modify) {
 			pmap_remove_write(m);
-			vfs_page_set_valid(bp, foff, m);
+			vfs_page_set_validclean(bp, foff, m);
 		} else if (m->valid == VM_PAGE_BITS_ALL &&
 		    (bp->b_flags & B_CACHE) == 0) {
 			bp->b_pages[i] = bogus_page;
@@ -3591,7 +3616,7 @@ vfs_clean_pages(struct buf *bp)
 
 		if (eoff > bp->b_offset + bp->b_bufsize)
 			eoff = bp->b_offset + bp->b_bufsize;
-		vfs_page_set_valid(bp, foff, m);
+		vfs_page_set_validclean(bp, foff, m);
 		/* vm_page_clear_dirty(m, foff & PAGE_MASK, eoff - foff); */
 		foff = noff;
 	}
diff --git a/sys/vm/vm_page.c b/sys/vm/vm_page.c
index 8befdd56705a..f0f90425a13a 100644
--- a/sys/vm/vm_page.c
+++ b/sys/vm/vm_page.c
@@ -1852,6 +1852,51 @@ vm_page_bits(int base, int size)
 }
 
 /*
+ *	vm_page_set_valid:
+ *
+ *	Sets portions of a page valid.  The arguments are expected
+ *	to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
+ *	of any partial chunks touched by the range.  The invalid portion of
+ *	such chunks will be zeroed.
+ *
+ *	(base + size) must be less then or equal to PAGE_SIZE.
+ */
+void
+vm_page_set_valid(vm_page_t m, int base, int size)
+{
+	int endoff, frag;
+
+	VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
+	if (size == 0)	/* handle degenerate case */
+		return;
+
+	/*
+	 * If the base is not DEV_BSIZE aligned and the valid
+	 * bit is clear, we have to zero out a portion of the
+	 * first block.
+	 */
+	if ((frag = base & ~(DEV_BSIZE - 1)) != base &&
+	    (m->valid & (1 << (base >> DEV_BSHIFT))) == 0)
+		pmap_zero_page_area(m, frag, base - frag);
+
+	/*
+	 * If the ending offset is not DEV_BSIZE aligned and the 
+	 * valid bit is clear, we have to zero out a portion of
+	 * the last block.
+	 */
+	endoff = base + size;
+	if ((frag = endoff & ~(DEV_BSIZE - 1)) != endoff &&
+	    (m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0)
+		pmap_zero_page_area(m, endoff,
+		    DEV_BSIZE - (endoff & (DEV_BSIZE - 1)));
+
+	/*
+	 * Set valid bits inclusive of any overlap.
+	 */
+	m->valid |= vm_page_bits(base, size);
+}
+
+/*
  *	vm_page_set_validclean:
  *
  *	Sets portions of a page valid and clean.  The arguments are expected
diff --git a/sys/vm/vm_page.h b/sys/vm/vm_page.h
index 7f996ea9bfab..8bbbaee192f4 100644
--- a/sys/vm/vm_page.h
+++ b/sys/vm/vm_page.h
@@ -321,6 +321,7 @@ vm_page_t vm_page_lookup (vm_object_t, vm_pindex_t);
 void vm_page_remove (vm_page_t);
 void vm_page_rename (vm_page_t, vm_object_t, vm_pindex_t);
 void vm_page_requeue(vm_page_t m);
+void vm_page_set_valid(vm_page_t m, int base, int size);
 void vm_page_sleep(vm_page_t m, const char *msg);
 vm_page_t vm_page_splay(vm_pindex_t, vm_page_t);
 vm_offset_t vm_page_startup(vm_offset_t vaddr);