1 files changed, 131 insertions, 26 deletions

diff --git a/usr/src/uts/common/fs/vnode.c b/usr/src/uts/common/fs/vnode.c
index 4abb040de0..9fb1ea702a 100644
--- a/usr/src/uts/common/fs/vnode.c
+++ b/usr/src/uts/common/fs/vnode.c
@@ -21,7 +21,7 @@
 
 /*
  * Copyright (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2013, Joyent, Inc. All rights reserved.
+ * Copyright (c) 2015, Joyent, Inc.
  */
 
 /*	Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T	*/
@@ -66,6 +66,7 @@
 #include <fs/fs_subr.h>
 #include <sys/taskq.h>
 #include <fs/fs_reparse.h>
+#include <sys/time.h>
 
 /* Determine if this vnode is a file that is read-only */
 #define	ISROFILE(vp)	\
@@ -102,6 +103,9 @@ kmutex_t	vskstat_tree_lock;
 /* Global variable which enables/disables the vopstats collection */
 int vopstats_enabled = 1;
 
+/* Global used for empty/invalid v_path */
+char *vn_vpath_empty = "";
+
 /*
  * forward declarations for internal vnode specific data (vsd)
  */
@@ -200,6 +204,11 @@ static void 		(**vsd_destructor)(void *);
 		cr = crgetmapped(cr);					\
 	}
 
+#define	VOP_LATENCY_10MS	10000000
+#define	VOP_LATENCY_100MS	100000000
+#define	VOP_LATENCY_1S		1000000000
+#define	VOP_LATENCY_10S		10000000000
+
 /*
  * Convert stat(2) formats to vnode types and vice versa.  (Knows about
  * numerical order of S_IFMT and vnode types.)
@@ -2284,7 +2293,7 @@ vn_cache_constructor(void *buf, void *cdrarg, int kmflags)
 	cv_init(&vp->v_cv, NULL, CV_DEFAULT, NULL);
 	rw_init(&vp->v_nbllock, NULL, RW_DEFAULT, NULL);
 	vp->v_femhead = NULL;	/* Must be done before vn_reinit() */
-	vp->v_path = NULL;
+	vp->v_path = vn_vpath_empty;
 	vp->v_mpssdata = NULL;
 	vp->v_vsd = NULL;
 	vp->v_fopdata = NULL;
@@ -2331,6 +2340,7 @@ void
 vn_recycle(vnode_t *vp)
 {
 	ASSERT(vp->v_pages == NULL);
+	VERIFY(vp->v_path != NULL);
 
 	/*
 	 * XXX - This really belongs in vn_reinit(), but we have some issues
@@ -2353,9 +2363,9 @@ vn_recycle(vnode_t *vp)
 		kmem_free(vp->v_femhead, sizeof (*(vp->v_femhead)));
 		vp->v_femhead = NULL;
 	}
-	if (vp->v_path) {
+	if (vp->v_path != vn_vpath_empty) {
 		kmem_free(vp->v_path, strlen(vp->v_path) + 1);
-		vp->v_path = NULL;
+		vp->v_path = vn_vpath_empty;
 	}
 
 	if (vp->v_fopdata != NULL) {
@@ -2427,9 +2437,10 @@ vn_free(vnode_t *vp)
 	 */
 	ASSERT((vp->v_count == 0) || (vp->v_count == 1));
 	ASSERT(vp->v_count_dnlc == 0);
-	if (vp->v_path != NULL) {
+	VERIFY(vp->v_path != NULL);
+	if (vp->v_path != vn_vpath_empty) {
 		kmem_free(vp->v_path, strlen(vp->v_path) + 1);
-		vp->v_path = NULL;
+		vp->v_path = vn_vpath_empty;
 	}
 
 	/* If FEM was in use, make sure everything gets cleaned up */
@@ -2516,6 +2527,7 @@ vnevent_rename_src(vnode_t *vp, vnode_t *dvp, char *name, caller_context_t *ct)
 	if (vp == NULL || vp->v_femhead == NULL) {
 		return;
 	}
+	(void) VOP_VNEVENT(dvp, VE_RENAME_SRC_DIR, vp, name, ct);
 	(void) VOP_VNEVENT(vp, VE_RENAME_SRC, dvp, name, ct);
 }
 
@@ -2530,12 +2542,13 @@ vnevent_rename_dest(vnode_t *vp, vnode_t *dvp, char *name,
 }
 
 void
-vnevent_rename_dest_dir(vnode_t *vp, caller_context_t *ct)
+vnevent_rename_dest_dir(vnode_t *vp, vnode_t *nvp, char *name,
+    caller_context_t *ct)
 {
 	if (vp == NULL || vp->v_femhead == NULL) {
 		return;
 	}
-	(void) VOP_VNEVENT(vp, VE_RENAME_DEST_DIR, NULL, NULL, ct);
+	(void) VOP_VNEVENT(vp, VE_RENAME_DEST_DIR, nvp, name, ct);
 }
 
 void
@@ -2951,7 +2964,7 @@ vn_setpath(vnode_t *rootvp, struct vnode *startvp, struct vnode *vp,
 	 * the potential for deadlock.
 	 */
 	mutex_enter(&base->v_lock);
-	if (base->v_path == NULL) {
+	if (base->v_path == vn_vpath_empty) {
 		mutex_exit(&base->v_lock);
 		return;
 	}
@@ -2978,7 +2991,8 @@ vn_setpath(vnode_t *rootvp, struct vnode *startvp, struct vnode *vp,
 	rpath = kmem_alloc(rpathalloc, KM_SLEEP);
 
 	mutex_enter(&base->v_lock);
-	if (base->v_path == NULL || strlen(base->v_path) != rpathlen) {
+	if (base->v_path == vn_vpath_empty ||
+	    strlen(base->v_path) != rpathlen) {
 		mutex_exit(&base->v_lock);
 		kmem_free(rpath, rpathalloc);
 		return;
@@ -2992,7 +3006,7 @@ vn_setpath(vnode_t *rootvp, struct vnode *startvp, struct vnode *vp,
 	rpath[rpathlen + plen] = '\0';
 
 	mutex_enter(&vp->v_lock);
-	if (vp->v_path != NULL) {
+	if (vp->v_path != vn_vpath_empty) {
 		mutex_exit(&vp->v_lock);
 		kmem_free(rpath, rpathalloc);
 	} else {
@@ -3012,7 +3026,7 @@ vn_setpath_str(struct vnode *vp, const char *str, size_t len)
 	char *buf = kmem_alloc(len + 1, KM_SLEEP);
 
 	mutex_enter(&vp->v_lock);
-	if (vp->v_path != NULL) {
+	if (vp->v_path != vn_vpath_empty) {
 		mutex_exit(&vp->v_lock);
 		kmem_free(buf, len + 1);
 		return;
@@ -3036,10 +3050,10 @@ vn_renamepath(vnode_t *dvp, vnode_t *vp, const char *nm, size_t len)
 
 	mutex_enter(&vp->v_lock);
 	tmp = vp->v_path;
-	vp->v_path = NULL;
+	vp->v_path = vn_vpath_empty;
 	mutex_exit(&vp->v_lock);
 	vn_setpath(rootdir, dvp, vp, nm, len);
-	if (tmp != NULL)
+	if (tmp != vn_vpath_empty)
 		kmem_free(tmp, strlen(tmp) + 1);
 }
 
@@ -3054,7 +3068,7 @@ vn_copypath(struct vnode *src, struct vnode *dst)
 	int alloc;
 
 	mutex_enter(&src->v_lock);
-	if (src->v_path == NULL) {
+	if (src->v_path == vn_vpath_empty) {
 		mutex_exit(&src->v_lock);
 		return;
 	}
@@ -3064,7 +3078,7 @@ vn_copypath(struct vnode *src, struct vnode *dst)
 	mutex_exit(&src->v_lock);
 	buf = kmem_alloc(alloc, KM_SLEEP);
 	mutex_enter(&src->v_lock);
-	if (src->v_path == NULL || strlen(src->v_path) + 1 != alloc) {
+	if (src->v_path == vn_vpath_empty || strlen(src->v_path) + 1 != alloc) {
 		mutex_exit(&src->v_lock);
 		kmem_free(buf, alloc);
 		return;
@@ -3073,7 +3087,7 @@ vn_copypath(struct vnode *src, struct vnode *dst)
 	mutex_exit(&src->v_lock);
 
 	mutex_enter(&dst->v_lock);
-	if (dst->v_path != NULL) {
+	if (dst->v_path != vn_vpath_empty) {
 		mutex_exit(&dst->v_lock);
 		kmem_free(buf, alloc);
 		return;
@@ -3231,14 +3245,57 @@ fop_read(
 	cred_t *cr,
 	caller_context_t *ct)
 {
-	int	err;
 	ssize_t	resid_start = uiop->uio_resid;
+	zone_t	*zonep = curzone;
+	zone_vfs_kstat_t *zvp = zonep->zone_vfs_stats;
+
+	hrtime_t start, lat;
+	ssize_t len;
+	int err;
+
+	if (vp->v_type == VREG || vp->v_type == VDIR || vp->v_type == VBLK) {
+		start = gethrtime();
+
+		mutex_enter(&zonep->zone_vfs_lock);
+		kstat_runq_enter(&zonep->zone_vfs_rwstats);
+		mutex_exit(&zonep->zone_vfs_lock);
+	}
 
 	VOPXID_MAP_CR(vp, cr);
 
 	err = (*(vp)->v_op->vop_read)(vp, uiop, ioflag, cr, ct);
-	VOPSTATS_UPDATE_IO(vp, read,
-	    read_bytes, (resid_start - uiop->uio_resid));
+	len = resid_start - uiop->uio_resid;
+
+	VOPSTATS_UPDATE_IO(vp, read, read_bytes, len);
+
+	if (vp->v_type == VREG || vp->v_type == VDIR || vp->v_type == VBLK) {
+		mutex_enter(&zonep->zone_vfs_lock);
+		zonep->zone_vfs_rwstats.reads++;
+		zonep->zone_vfs_rwstats.nread += len;
+		kstat_runq_exit(&zonep->zone_vfs_rwstats);
+		mutex_exit(&zonep->zone_vfs_lock);
+
+		lat = gethrtime() - start;
+
+		if (lat >= VOP_LATENCY_10MS) {
+			if (lat < VOP_LATENCY_100MS)
+				atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
+			else if (lat < VOP_LATENCY_1S) {
+				atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
+				atomic_inc_64(&zvp->zv_100ms_ops.value.ui64);
+			} else if (lat < VOP_LATENCY_10S) {
+				atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
+				atomic_inc_64(&zvp->zv_100ms_ops.value.ui64);
+				atomic_inc_64(&zvp->zv_1s_ops.value.ui64);
+			} else {
+				atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
+				atomic_inc_64(&zvp->zv_100ms_ops.value.ui64);
+				atomic_inc_64(&zvp->zv_1s_ops.value.ui64);
+				atomic_inc_64(&zvp->zv_10s_ops.value.ui64);
+			}
+		}
+	}
+
 	return (err);
 }
 
@@ -3250,14 +3307,62 @@ fop_write(
 	cred_t *cr,
 	caller_context_t *ct)
 {
-	int	err;
 	ssize_t	resid_start = uiop->uio_resid;
+	zone_t	*zonep = curzone;
+	zone_vfs_kstat_t *zvp = zonep->zone_vfs_stats;
+
+	hrtime_t start, lat;
+	ssize_t len;
+	int	err;
+
+	/*
+	 * For the purposes of VFS kstat consumers, the "waitq" calculation is
+	 * repurposed as the active queue for VFS write operations.  There's no
+	 * actual wait queue for VFS operations.
+	 */
+	if (vp->v_type == VREG || vp->v_type == VDIR || vp->v_type == VBLK) {
+		start = gethrtime();
+
+		mutex_enter(&zonep->zone_vfs_lock);
+		kstat_waitq_enter(&zonep->zone_vfs_rwstats);
+		mutex_exit(&zonep->zone_vfs_lock);
+	}
 
 	VOPXID_MAP_CR(vp, cr);
 
 	err = (*(vp)->v_op->vop_write)(vp, uiop, ioflag, cr, ct);
-	VOPSTATS_UPDATE_IO(vp, write,
-	    write_bytes, (resid_start - uiop->uio_resid));
+	len = resid_start - uiop->uio_resid;
+
+	VOPSTATS_UPDATE_IO(vp, write, write_bytes, len);
+
+	if (vp->v_type == VREG || vp->v_type == VDIR || vp->v_type == VBLK) {
+		mutex_enter(&zonep->zone_vfs_lock);
+		zonep->zone_vfs_rwstats.writes++;
+		zonep->zone_vfs_rwstats.nwritten += len;
+		kstat_waitq_exit(&zonep->zone_vfs_rwstats);
+		mutex_exit(&zonep->zone_vfs_lock);
+
+		lat = gethrtime() - start;
+
+		if (lat >= VOP_LATENCY_10MS) {
+			if (lat < VOP_LATENCY_100MS)
+				atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
+			else if (lat < VOP_LATENCY_1S) {
+				atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
+				atomic_inc_64(&zvp->zv_100ms_ops.value.ui64);
+			} else if (lat < VOP_LATENCY_10S) {
+				atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
+				atomic_inc_64(&zvp->zv_100ms_ops.value.ui64);
+				atomic_inc_64(&zvp->zv_1s_ops.value.ui64);
+			} else {
+				atomic_inc_64(&zvp->zv_10ms_ops.value.ui64);
+				atomic_inc_64(&zvp->zv_100ms_ops.value.ui64);
+				atomic_inc_64(&zvp->zv_1s_ops.value.ui64);
+				atomic_inc_64(&zvp->zv_10s_ops.value.ui64);
+			}
+		}
+	}
+
 	return (err);
 }
 
@@ -3421,7 +3526,7 @@ fop_lookup(
 	}
 	if (ret == 0 && *vpp) {
 		VOPSTATS_UPDATE(*vpp, lookup);
-		if ((*vpp)->v_path == NULL) {
+		if ((*vpp)->v_path == vn_vpath_empty) {
 			vn_setpath(rootdir, dvp, *vpp, nm, strlen(nm));
 		}
 	}
@@ -3463,7 +3568,7 @@ fop_create(
 	    (dvp, name, vap, excl, mode, vpp, cr, flags, ct, vsecp);
 	if (ret == 0 && *vpp) {
 		VOPSTATS_UPDATE(*vpp, create);
-		if ((*vpp)->v_path == NULL) {
+		if ((*vpp)->v_path == vn_vpath_empty) {
 			vn_setpath(rootdir, dvp, *vpp, name, strlen(name));
 		}
 	}
@@ -3585,7 +3690,7 @@ fop_mkdir(
 	    (dvp, dirname, vap, vpp, cr, ct, flags, vsecp);
 	if (ret == 0 && *vpp) {
 		VOPSTATS_UPDATE(*vpp, mkdir);
-		if ((*vpp)->v_path == NULL) {
+		if ((*vpp)->v_path == vn_vpath_empty) {
 			vn_setpath(rootdir, dvp, *vpp, dirname,
 			    strlen(dirname));
 		}