Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S270408AbTGRVle (ORCPT ); Fri, 18 Jul 2003 17:41:34 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S271880AbTGRVku (ORCPT ); Fri, 18 Jul 2003 17:40:50 -0400 Received: from inet-mail1.oracle.com ([148.87.2.201]:6836 "EHLO inet-mail1.oracle.com") by vger.kernel.org with ESMTP id S271824AbTGRVjk (ORCPT ); Fri, 18 Jul 2003 17:39:40 -0400 Date: Fri, 18 Jul 2003 14:54:22 -0700 From: Mark Fasheh To: marcelo@conectiva.com.br Cc: sct@redhat.com, linux-kernel@vger.kernel.org Subject: [PATCH-2.4] [RESEND] Fix deadlock in journal_create Message-ID: <20030718215421.GF1014@ca-server1.us.oracle.com> Reply-To: Mark Fasheh Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline Organization: Oracle Corporation User-Agent: Mutt/1.5.4i Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 2023 Lines: 55 Marcelo, I sent this initially against 2.4.21-rc6 and it didn't make it in -- even though it got Stephen's OK. Here's a resend -- I've identified that the bug still exists and 2.4.22-pre7. The patch didn't need to be changed as it still applies cleanly. I ran across a deadlock when trying to do a journal_create at mount time. The problem is that journal_create does a sync_dev which eventually tries to do a get_super which does: down_read(&s->s_umount); The problem arises if I call journal_create from my read_super method in which case get_sb_bdev has already done: down_write(&s->s_umount); from alloc_super. Replacing the sync_dev call with an fsync_no_super seems to have fixed the deadlock. It you want to test this out using ext3 (I have verified it with my own filesystem), simply follow these steps: 1) create an ext2 filesystem on a device 2) mount that new partition and make a "journal file" on it (using dd). get the inode number of that file. 3) unmount it and remount it as an ext3 filesystem using the option: journal=inode_number where inode number is the inode number of the journal file you just created. This will get ext3's read_super method to call ext3_create_journal which will hang during a journal_create. A trivial patch to fix this is attached. --Mark -- Mark Fasheh Software Developer, Oracle Corp mark.fasheh@oracle.com --- linux-2.4.22-pre7/fs/jbd/journal.c.orig 2003-05-30 11:49:10.000000000 -0700 +++ linux-2.4.22-pre7/fs/jbd/journal.c 2003-05-30 11:49:18.000000000 -0700 @@ -912,7 +912,7 @@ int journal_create(journal_t *journal) __brelse(bh); } - sync_dev(journal->j_dev); + fsync_no_super(journal->j_dev); jbd_debug(1, "JBD: journal cleared.\n"); /* OK, fill in the initial static fields in the new superblock */ - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/