From owner-freebsd-current@FreeBSD.ORG Tue Jun 17 12:58:24 2003 Return-Path: Delivered-To: freebsd-current@freebsd.org Received: from mx1.FreeBSD.org (mx1.freebsd.org [216.136.204.125]) by hub.freebsd.org (Postfix) with ESMTP id B137337B401; Tue, 17 Jun 2003 12:58:24 -0700 (PDT) Received: from fledge.watson.org (fledge.watson.org [204.156.12.50]) by mx1.FreeBSD.org (Postfix) with ESMTP id F20B543F85; Tue, 17 Jun 2003 12:58:23 -0700 (PDT) (envelope-from robert@fledge.watson.org) Received: from fledge.watson.org (localhost [127.0.0.1]) by fledge.watson.org (8.12.9/8.12.9) with ESMTP id h5HJudYA018042; Tue, 17 Jun 2003 15:56:39 -0400 (EDT) (envelope-from robert@fledge.watson.org) Received: from localhost (robert@localhost)h5HJudVD018039; Tue, 17 Jun 2003 15:56:39 -0400 (EDT) (envelope-from robert@fledge.watson.org) Date: Tue, 17 Jun 2003 15:56:39 -0400 (EDT) From: Robert Watson X-Sender: robert@fledge.watson.org To: Don Lewis In-Reply-To: <200306171044.h5HAigM7051410@gw.catspoiler.org> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII cc: current@FreeBSD.org Subject: Re: fdrop_locked() and FILE_LOCK() vs. Giant X-BeenThere: freebsd-current@freebsd.org X-Mailman-Version: 2.1.1 Precedence: list List-Id: Discussions about the use of FreeBSD-current List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Tue, 17 Jun 2003 19:58:25 -0000 On Tue, 17 Jun 2003, Don Lewis wrote: > The FILE_LOCK() implementation uses "pool mutex" under the hood, which > means it should only be used as a leaf level mutex. The fdrop_locked() > code wants to be called with FILE_LOCK() held, but the fdrop_locked() > implementation calls mtx_lock(&Giant) before calling FILE_UNLOCK(). In > addition to violating the proper usage of the "pool mutex", there is > also the potential for a lock order violation. The close() > implementation grabs Giant and eventually calls fdrop(), which calls > FILE_LOCK() immediately before calling fdrop_locked(). If another > caller of fdrop_locked() calls FILE_LOCK() without grabbing Giant first, > then the lock order will be reversed when fdrop_locked() grabs Giant. > > It looks like fdrop_locked() should require that Giant be grabbed by the > caller before fdrop_locked() is called. I've also noticed that the file descriptor lock is held over per-object calls to fo_poll(), which currently isn't a big deal for most objects, but may be in the future if we have to grab other locks in order to test the poll status inside the object. I run into this problem with the MAC work because the vnode label is protected by the vnode lock, which is a sleepable lock. We may want to change label locking in the future to avoid this, but in the mean time I get a lot of witness warnings, and using a pool mutex for the fd lock may cause lock order problems if there is more complex locking. Robert N M Watson FreeBSD Core Team, TrustedBSD Projects robert@fledge.watson.org Network Associates Laboratories