From owner-svn-src-stable@freebsd.org Sun Nov 6 17:17:53 2016 Return-Path: Delivered-To: svn-src-stable@mailman.ysv.freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2001:1900:2254:206a::19:1]) by mailman.ysv.freebsd.org (Postfix) with ESMTP id E925FC337B8; Sun, 6 Nov 2016 17:17:53 +0000 (UTC) (envelope-from alc@FreeBSD.org) Received: from repo.freebsd.org (repo.freebsd.org [IPv6:2610:1c1:1:6068::e6a:0]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (Client did not present a certificate) by mx1.freebsd.org (Postfix) with ESMTPS id A6E99F7F; Sun, 6 Nov 2016 17:17:53 +0000 (UTC) (envelope-from alc@FreeBSD.org) Received: from repo.freebsd.org ([127.0.1.37]) by repo.freebsd.org (8.15.2/8.15.2) with ESMTP id uA6HHq6T069376; Sun, 6 Nov 2016 17:17:52 GMT (envelope-from alc@FreeBSD.org) Received: (from alc@localhost) by repo.freebsd.org (8.15.2/8.15.2/Submit) id uA6HHqM8069375; Sun, 6 Nov 2016 17:17:52 GMT (envelope-from alc@FreeBSD.org) Message-Id: <201611061717.uA6HHqM8069375@repo.freebsd.org> X-Authentication-Warning: repo.freebsd.org: alc set sender to alc@FreeBSD.org using -f From: Alan Cox Date: Sun, 6 Nov 2016 17:17:52 +0000 (UTC) To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-stable@freebsd.org, svn-src-stable-11@freebsd.org Subject: svn commit: r308378 - stable/11/sys/vm X-SVN-Group: stable-11 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit X-BeenThere: svn-src-stable@freebsd.org X-Mailman-Version: 2.1.23 Precedence: list List-Id: SVN commit messages for all the -stable branches of the src tree List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Sun, 06 Nov 2016 17:17:54 -0000 Author: alc Date: Sun Nov 6 17:17:52 2016 New Revision: 308378 URL: https://svnweb.freebsd.org/changeset/base/308378 Log: MFC r308174, r308261 Move and revise a comment about the relation between the object's paging- in-progress count and the vnode. Prior to r188331, we always acquired the vnode lock before incrementing the object's paging-in-progress count. Now, we increment it before attempting to acquire the vnode lock with LK_NOWAIT, but we never sleep acquiring the vnode lock while we have the count incremented. In vm_fault()'s loop over the shadow chain, move a comment describing our invariants to a better place. Also, add two comments concerning the relationship between the map and vnode locks. Modified: stable/11/sys/vm/vm_fault.c Directory Properties: stable/11/ (props changed) Modified: stable/11/sys/vm/vm_fault.c ============================================================================== --- stable/11/sys/vm/vm_fault.c Sun Nov 6 17:12:02 2016 (r308377) +++ stable/11/sys/vm/vm_fault.c Sun Nov 6 17:17:52 2016 (r308378) @@ -420,8 +420,7 @@ fast_failed: * they will stay around as well. * * Bump the paging-in-progress count to prevent size changes (e.g. - * truncation operations) during I/O. This must be done after - * obtaining the vnode lock in order to avoid possible deadlocks. + * truncation operations) during I/O. */ vm_object_reference_locked(fs.first_object); vm_object_pip_add(fs.first_object, 1); @@ -565,6 +564,14 @@ fast_failed: readrest: /* + * At this point, we have either allocated a new page or found + * an existing page that is only partially valid. + * + * We hold a reference on the current object and the page is + * exclusive busied. + */ + + /* * If the pager for the current object might have the page, * then determine the number of additional pages to read and * potentially reprioritize previously read pages for earlier @@ -633,21 +640,32 @@ readrest: */ if (fs.object->type != OBJT_DEFAULT) { /* - * We have either allocated a new page or found an - * existing page that is only partially valid. We - * hold a reference on fs.object and the page is - * exclusive busied. + * Release the map lock before locking the vnode or + * sleeping in the pager. (If the current object has + * a shadow, then an earlier iteration of this loop + * may have already unlocked the map.) */ unlock_map(&fs); if (fs.object->type == OBJT_VNODE && (vp = fs.object->handle) != fs.vp) { + /* + * Perform an unlock in case the desired vnode + * changed while the map was unlocked during a + * retry. + */ unlock_vp(&fs); - locked = VOP_ISLOCKED(vp); + locked = VOP_ISLOCKED(vp); if (locked != LK_EXCLUSIVE) locked = LK_SHARED; - /* Do not sleep for vnode lock while fs.m is busy */ + + /* + * We must not sleep acquiring the vnode lock + * while we have the page exclusive busied or + * the object's paging-in-progress count + * incremented. Otherwise, we could deadlock. + */ error = vget(vp, locked | LK_CANRECURSE | LK_NOWAIT, curthread); if (error != 0) {