From mboxrd@z Thu Jan 1 00:00:00 1970 From: Edward Shishkin Subject: Re: 2.6.23-rc1-mm2 Invalid argument on some files, depends on file contents? Date: Thu, 02 Aug 2007 03:48:44 +0400 Message-ID: <46B11BDC.8050608@namesys.com> References: <1186005421.9066.9.camel@localhost> Mime-Version: 1.0 Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <1186005421.9066.9.camel@localhost> Sender: reiserfs-devel-owner@vger.kernel.org List-Id: Content-Type: text/plain; charset="us-ascii"; format="flowed" To: Zan Lynx Cc: ReiserFS Mailing List Zan Lynx wrote: >Several of my files have been doing this. I really noticed it when I >ran a backup last night and there were many of these errors. It has >also started affecting documents I really care about. > >I was wondering if I somehow have compression enabled. I didn't do it >on purpose! :) > > > If so, then definitely _no_. >I remember reading some comment about lzo being messed up and the >patches not included in mm yet? This is rc1-mm2 so I thought it might >be fixed, but I guess not. > > Reiser4 in rc1-mm2 is already ported to the new lzo, however this library is not used by default (unix-file) plugin. >Here is a snippet of a shell session showing the problem. It's a good >thing it didn't affect the OpenOffice format original! > >$ cat * >/dev/null >cat: URL Database Design.doc.backup: Invalid argument >$ cp URL\ Database\ Design.doc.backup{,.test} >$ cat * >/dev/null >cat: URL Database Design.doc.backup: Invalid argument >cat: URL Database Design.doc.backup.test: Invalid argument >$ echo blah >> URL\ Database\ Design.doc.backup.test >$ cat * >/dev/null >cat: URL Database Design.doc.backup: Invalid argument >$ ls -l >-rw-r--r-- 1 lynx users 286720 2007-07-31 18:16 URL Database Design.doc.backup >-rw-r--r-- 1 lynx users 286725 2007-08-01 15:48 URL Database Design.doc.backup.test >-rw-r--r-- 1 lynx users 154915 2007-07-31 16:53 URL Database Design.odt >-rw-r--r-- 1 lynx users 154915 2007-07-31 18:16 URL Database Design.odt.backup > >As you can see, a copy shows the same problem. But adding "blah\n" to >the end makes the problem go away... > > Yeah, I have also noted this badness: if file size % page size == 0, then md5sum complaints. This is one more bug in new file_read: if vfs asks to read page out of file size, then fs should fill it by zeros (reiser4 returns -EINVAL instead). I'll prepare the patch a bit later.. Thanks, Edward.