- in genfs_getpages, take g_glock earlier so that it can't be

intervened by truncation.
  it also fixes a deadlock.  (g_glock vs pages locking order)
- uvm_vnp_setsize: modify v_size while holding v_interlock.

reviewed by Chuck Silvers.

sys/miscfs/genfs/genfs_vnops.c

@@ -1,4 +1,4 @@
-/*	$NetBSD: genfs_vnops.c,v 1.125 2006/05/14 21:31:52 elad Exp $	*/
+/*	$NetBSD: genfs_vnops.c,v 1.126 2006/07/22 08:47:56 yamt Exp $	*/
 
 /*
  * Copyright (c) 1982, 1986, 1989, 1993
@@ -31,7 +31,7 @@
  */
 
 #include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: genfs_vnops.c,v 1.125 2006/05/14 21:31:52 elad Exp $");
+__KERNEL_RCSID(0, "$NetBSD: genfs_vnops.c,v 1.126 2006/07/22 08:47:56 yamt Exp $");
 
 #if defined(_KERNEL_OPT)
 #include "opt_nfsserver.h"
@@ -460,6 +460,7 @@ genfs_getpages(void *v)
 	boolean_t sawhole = FALSE;
 	boolean_t overwrite = (flags & PGO_OVERWRITE) != 0;
 	boolean_t blockalloc = write && (flags & PGO_NOBLOCKALLOC) == 0;
+	voff_t origvsize;
 	UVMHIST_FUNC("genfs_getpages"); UVMHIST_CALLED(ubchist);
 
 	UVMHIST_LOG(ubchist, "vp %p off 0x%x/%x count %d",
@@ -473,7 +474,9 @@ genfs_getpages(void *v)
 		panic("genfs_getpages: too many pages");
 	}
 
+startover:
 	error = 0;
+	origvsize = vp->v_size;
 	origoffset = ap->a_offset;
 	orignpages = *ap->a_count;
 	GOP_SIZE(vp, vp->v_size, &diskeof, 0);
@@ -539,6 +542,7 @@ genfs_getpages(void *v)
 
 		return (ap->a_m[ap->a_centeridx] == NULL ? EBUSY : 0);
 	}
+	simple_unlock(&uobj->vmobjlock);
 
 	/*
 	 * find the requested pages and make some simple checks.
@@ -566,9 +570,9 @@ genfs_getpages(void *v)
 	pgs_size = sizeof(struct vm_page *) *
 	    ((endoffset - startoffset) >> PAGE_SHIFT);
 	if (pgs_size > sizeof(pgs_onstack)) {
-		pgs = malloc(pgs_size, M_DEVBUF, M_NOWAIT | M_ZERO);
+		pgs = malloc(pgs_size, M_DEVBUF,
+		    (async ? M_NOWAIT : M_WAITOK) | M_ZERO);
 		if (pgs == NULL) {
-			simple_unlock(&uobj->vmobjlock);
 			return (ENOMEM);
 		}
 	} else {
@@ -577,8 +581,29 @@ genfs_getpages(void *v)
 	}
 	UVMHIST_LOG(ubchist, "ridx %d npages %d startoff %ld endoff %ld",
 	    ridx, npages, startoffset, endoffset);
+
+	/*
+	 * hold g_glock to prevent a race with truncate.
+	 *
+	 * check if our idea of v_size is still valid.
+	 */
+
+	if (blockalloc) {
+		lockmgr(&gp->g_glock, LK_EXCLUSIVE, NULL);
+	} else {
+		lockmgr(&gp->g_glock, LK_SHARED, NULL);
+	}
+	simple_lock(&uobj->vmobjlock);
+	if (vp->v_size < origvsize) {
+		lockmgr(&gp->g_glock, LK_RELEASE, NULL);
+		if (pgs != pgs_onstack)
+			free(pgs, M_DEVBUF);
+		goto startover;
+	}
+
 	if (uvn_findpages(uobj, origoffset, &npages, &pgs[ridx],
 	    async ? UFP_NOWAIT : UFP_ALL) != orignpages) {
+		lockmgr(&gp->g_glock, LK_RELEASE, NULL);
 		KASSERT(async != 0);
 		genfs_rel_pages(&pgs[ridx], orignpages);
 		simple_unlock(&uobj->vmobjlock);
@@ -600,6 +625,7 @@ genfs_getpages(void *v)
 		}
 	}
 	if (i == npages) {
+		lockmgr(&gp->g_glock, LK_RELEASE, NULL);
 		UVMHIST_LOG(ubchist, "returning cached pages", 0,0,0,0);
 		npages += ridx;
 		goto out;
@@ -610,6 +636,7 @@ genfs_getpages(void *v)
 	 */
 
 	if (overwrite) {
+		lockmgr(&gp->g_glock, LK_RELEASE, NULL);
 		UVMHIST_LOG(ubchist, "PGO_OVERWRITE",0,0,0,0);
 
 		for (i = 0; i < npages; i++) {
@@ -644,6 +671,7 @@ genfs_getpages(void *v)
 		npgs = npages;
 		if (uvn_findpages(uobj, startoffset, &npgs, pgs,
 		    async ? UFP_NOWAIT : UFP_ALL) != npages) {
+			lockmgr(&gp->g_glock, LK_RELEASE, NULL);
 			KASSERT(async != 0);
 			genfs_rel_pages(pgs, npages);
 			simple_unlock(&uobj->vmobjlock);
@@ -700,12 +728,6 @@ genfs_getpages(void *v)
 	 * now loop over the pages, reading as needed.
 	 */
 
-	if (blockalloc) {
-		lockmgr(&gp->g_glock, LK_EXCLUSIVE, NULL);
-	} else {
-		lockmgr(&gp->g_glock, LK_SHARED, NULL);
-	}
-
 	bp = NULL;
 	for (offset = startoffset;
 	    bytes > 0;

sys/uvm/uvm_vnode.c

@@ -1,4 +1,4 @@
-/*	$NetBSD: uvm_vnode.c,v 1.71 2006/07/21 16:48:54 ad Exp $	*/
+/*	$NetBSD: uvm_vnode.c,v 1.72 2006/07/22 08:47:56 yamt Exp $	*/
 
 /*
  * Copyright (c) 1997 Charles D. Cranor and Washington University.
@@ -50,7 +50,7 @@
  */
 
 #include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: uvm_vnode.c,v 1.71 2006/07/21 16:48:54 ad Exp $");
+__KERNEL_RCSID(0, "$NetBSD: uvm_vnode.c,v 1.72 2006/07/22 08:47:56 yamt Exp $");
 
 #include "fs_nfs.h"
 #include "opt_uvmhist.h"
@@ -482,6 +482,7 @@ uvm_vnp_setsize(struct vnode *vp, voff_t newsize)
 {
 	struct uvm_object *uobj = &vp->v_uobj;
 	voff_t pgend = round_page(newsize);
+	voff_t oldsize;
 	UVMHIST_FUNC("uvm_vnp_setsize"); UVMHIST_CALLED(ubchist);
 
 	simple_lock(&uobj->vmobjlock);
@@ -493,12 +494,13 @@ uvm_vnp_setsize(struct vnode *vp, voff_t newsize)
 	 * toss some pages...
 	 */
 
-	if (vp->v_size > pgend && vp->v_size != VSIZENOTSET) {
+	oldsize = vp->v_size;
+	vp->v_size = newsize;
+	if (oldsize > pgend && oldsize != VSIZENOTSET) {
 		(void) uvn_put(uobj, pgend, 0, PGO_FREE | PGO_SYNCIO);
 	} else {
 		simple_unlock(&uobj->vmobjlock);
 	}
-	vp->v_size = newsize;
 }
 
 /*