 @@ -1,14 +1,14 @@
-/*	$NetBSD: vfs_subr.c,v 1.357.4.4 2009/02/16 03:33:16 snj Exp $	*/
+/*	$NetBSD: vfs_subr.c,v 1.357.4.5 2009/07/21 00:31:58 snj Exp $	*/
 /*-
  * Copyright (c) 1997, 1998, 2004, 2005, 2007, 2008 The NetBSD Foundation, Inc.
  * All rights reserved.
+ *
  * This code is derived from software contributed to The NetBSD Foundation
  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
  * NASA Ames Research Center, by Charles M. Hannum, and by Andrew Doran.
+ *
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions
  * are met:
  * 1. Redistributions of source code must retain the above copyright
 @@ -66,32 +66,42 @@
  *	@(#)vfs_subr.c	8.13 (Berkeley) 4/18/94
  */
 /*
  * Note on v_usecount and locking:
+ *
  * At nearly all points it is known that v_usecount could be zero, the
  * vnode interlock will be held.
+ *
  * To change v_usecount away from zero, the interlock must be held.  To
  * change from a non-zero value to zero, again the interlock must be
  * held.
+ *
- * Changing the usecount from a non-zero value to a non-zero value can
+ * There's a flag bit, VC_XLOCK, embedded in v_usecount.
- * safely be done using atomic operations, without the interlock held.
+ * To raise v_usecount, if the VC_XLOCK bit is set in it, the interlock
  * must be held.
  * To modify the VC_XLOCK bit, the interlock must be held.
  * We always keep the usecount (v_usecount & VC_MASK) non-zero while the
  * VC_XLOCK bit is set.
+ *
  * Unless the VC_XLOCK bit is set, changing the usecount from a non-zero
  * value to a non-zero value can safely be done using atomic operations,
  * without the interlock held.
  * Even if the VC_XLOCK bit is set, decreasing the usecount to a non-zero
  * value can be done using atomic operations, without the interlock held.
  */
 #include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: vfs_subr.c,v 1.357.4.4 2009/02/16 03:33:16 snj Exp $");
+__KERNEL_RCSID(0, "$NetBSD: vfs_subr.c,v 1.357.4.5 2009/07/21 00:31:58 snj Exp $");
 #include "opt_ddb.h"
 #include "opt_compat_netbsd.h"
 #include "opt_compat_43.h"
 #include <sys/param.h>
 #include <sys/systm.h>
 #include <sys/proc.h>
 #include <sys/kernel.h>
 #include <sys/mount.h>
 #include <sys/fcntl.h>
 #include <sys/vnode.h>
 #include <sys/stat.h>
 @@ -356,28 +366,30 @@ try_nextlist:
+	}
 	/* Remove it from the freelist. */
 	TAILQ_REMOVE(listhd, vp, v_freelist);
 	vp->v_freelisthd = NULL;
 	mutex_exit(&vnode_free_list_lock);
 	/*
 	 * The vnode is still associated with a file system, so we must
 	 * clean it out before reusing it.  We need to add a reference
 	 * before doing this.  If the vnode gains another reference while
 	 * being cleaned out then we lose - retry.
 	 */
-	atomic_inc_uint(&vp->v_usecount);
+	atomic_add_int(&vp->v_usecount, 1 + VC_XLOCK);
 	vclean(vp, DOCLOSE);
 	KASSERT(vp->v_usecount >= 1 + VC_XLOCK);
 	atomic_add_int(&vp->v_usecount, -VC_XLOCK);
 	if (vp->v_usecount == 1) {
 		/* We're about to dirty it. */
 		vp->v_iflag &= ~VI_CLEAN;
 		mutex_exit(&vp->v_interlock);
 		if (vp->v_type == VBLK || vp->v_type == VCHR) {
 			spec_node_destroy(vp);
+		}
 		vp->v_type = VNON;
 	} else {
 		/*
 		 * Don't return to freelist - the holder of the last
 		 * reference will destroy it.
 		 */
 @@ -1211,27 +1223,27 @@ bool
 vtryget(vnode_t *vp)
+{
 	u_int use, next;
 	/*
 	 * If the vnode is being freed, don't make life any harder
 	 * for vclean() by adding another reference without waiting.
 	 * This is not strictly necessary, but we'll do it anyway.
 	 */
 	if (__predict_false((vp->v_iflag & (VI_XLOCK | VI_FREEING)) != 0)) {
 		return false;
+	}
 	for (use = vp->v_usecount;; use = next) {
-		if (use == 0) {
+		if (use == 0 || __predict_false((use & VC_XLOCK) != 0)) {
 			/* Need interlock held if first reference. */
 			return false;
+		}
 		next = atomic_cas_uint(&vp->v_usecount, use, use + 1);
 		if (__predict_true(next == use)) {
 			return true;
+		}
+	}
+}
 /*
  * Grab a particular vnode from the free list, increment its
  * reference count and lock it. If the vnode lock bit is set the
 @@ -1300,29 +1312,30 @@ vput(vnode_t *vp)
 	vrele(vp);
+}
 /*
  * Try to drop reference on a vnode.  Abort if we are releasing the
  * last reference.  Note: this _must_ succeed if not the last reference.
  */
 static inline bool
 vtryrele(vnode_t *vp)
+{
 	u_int use, next;
 	for (use = vp->v_usecount;; use = next) {
 		if (use == 1) {
 			return false;
+		}
 		KASSERT((use & VC_MASK) > 1);
 		next = atomic_cas_uint(&vp->v_usecount, use, use - 1);
 		if (__predict_true(next == use)) {
 			return true;
+		}
+	}
+}
 /*
  * Vnode release.  If reference count drops to zero, call inactive
  * routine and either return to freelist or free to the pool.
  */
 void
 vrelel(vnode_t *vp, int flags)

 @@ -1,14 +1,14 @@
-/*	$NetBSD: vnode.h,v 1.197 2008/07/31 05:38:06 simonb Exp $	*/
+/*	$NetBSD: vnode.h,v 1.197.4.1 2009/07/21 00:31:58 snj Exp $	*/
 /*-
  * Copyright (c) 2008 The NetBSD Foundation, Inc.
  * All rights reserved.
+ *
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions
  * are met:
  * 1. Redistributions of source code must retain the above copyright
  *    notice, this list of conditions and the following disclaimer.
  * 2. Redistributions in binary form must reproduce the above copyright
  *    notice, this list of conditions and the following disclaimer in the
  *    documentation and/or other materials provided with the distribution.
 @@ -239,26 +239,32 @@ typedef struct vnode vnode_t;
  */
 #define	VU_DIROP	0x01000000	/* LFS: involved in a directory op */
 #define	VU_SOFTDEP	0x02000000	/* FFS: involved in softdep processing */
 #define	VNODE_FLAGBITS \
     "\20\1ROOT\2SYSTEM\3ISTTY\4MAPPED\5MPSAFE\6LOCKSWORK\11TEXT\12EXECMAP" \
     "\13WRMAP\14WRMAPDIRTY\15XLOCK\17ONWORKLST\20MARKER" \
     "\22LAYER\24CLEAN\25INACTPEND\26INACTREDO\27FREEING" \
     "\28INACTNOW\31DIROP\32SOFTDEP"
 #define	VSIZENOTSET	((voff_t)-1)
 /*
  * v_usecount; see the comment in vfs_subr.c
  */
 #define	VC_XLOCK	0x80000000
 #define	VC_MASK		0x7fffffff
 /*
  * Vnode attributes.  A field value of VNOVAL represents a field whose value
  * is unavailable (getattr) or which is not to be changed (setattr).
  */
 struct vattr {
 	enum vtype	va_type;	/* vnode type (for create) */
 	mode_t		va_mode;	/* files access mode and type */
 	nlink_t		va_nlink;	/* number of references to file */
 	uid_t		va_uid;		/* owner user id */
 	gid_t		va_gid;		/* owner group id */
 	long		va_fsid;	/* file system id (dev for now) */
 	ino_t		va_fileid;	/* file id */
 	u_quad_t	va_size;	/* file size in bytes */
 	long		va_blocksize;	/* blocksize preferred for i/o */