From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: with ECARTIS (v1.0.0; list xfs); Thu, 08 Nov 2007 21:53:56 -0800 (PST) Received: from larry.melbourne.sgi.com (larry.melbourne.sgi.com [134.14.52.130]) by oss.sgi.com (8.12.11.20060308/8.12.10/SuSE Linux 0.7) with SMTP id lA95rnXx021081 for ; Thu, 8 Nov 2007 21:53:51 -0800 Message-ID: <4733F301.9020706@sgi.com> Date: Fri, 09 Nov 2007 16:41:21 +1100 From: Timothy Shimmin MIME-Version: 1.0 Subject: Re: acl and attr: Fix path walking code References: <200710281858.24428.agruen@suse.de> In-Reply-To: <200710281858.24428.agruen@suse.de> Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Sender: xfs-bounce@oss.sgi.com Errors-to: xfs-bounce@oss.sgi.com List-Id: xfs To: Andreas Gruenbacher Cc: linux-xfs@oss.sgi.com, Gerald Bringhurst , Brandon Philips Hi Andreas, Andreas Gruenbacher wrote: > Hello, > > the tree walking code in acl and attr broke when resolve_symlinks() was > introduced (by me, unfortunately). Following symlinks passed in on the > command line is the intended behavior for the tools (unless in -P mode). The > first version was buggy, and so someone "fixed" it by replacing readlink() > with realpath() in resolve_symlinks(). > > The result is that the output of getfattr and getfacl will show pathnames that > may point anywhere. When processing a directory tree it sometimes is helpful > to treat symlinks as regular files, but resolving the pathnames is totally > wrong. > > After runnig into problem after problem with nftw and never ending up with > even half-way clean code, I think it's time to ditch it altogether and > replace it with sane code. So here are two patches, one for attr and one for > acl, that does that. > > Files include/walk_tree.h and libmisc/walk_tree.c are identical in both > patches; that code is shared between the two packages. > > Okay to apply? > > Thanks, > Andreas > I applied attr patch and tried it out on xfstests/062 (which I believe was based on one of your tests). ========================================================== --- 062.out 2006-03-28 12:52:32.000000000 +1000 +++ 062.out.bad 2007-11-09 15:38:09.000000000 +1100 @@ -526,6 +526,10 @@ user.name=0xbabe user.name3=0xdeface +# file: SCRATCH_MNT/lnk +trusted.name=0xbabe +trusted.name3=0xdeface + # file: SCRATCH_MNT/dev/b trusted.name=0xbabe trusted.name3=0xdeface @@ -562,6 +566,10 @@ user.1=0x3233 user.x=0x797a +# file: SCRATCH_MNT/descend/and/ascend +trusted.9=0x3837 +trusted.a=0x6263 + *** directory descent without following symlinks # file: SCRATCH_MNT/reg ========================================================== So for the following of symlinks with getfattr -L i.e. echo "*** directory descent with us following symlinks" getfattr -h -L -R -m '.' -e hex $SCRATCH_MNT Looking at the 2nd difference... It now picks up descend/and/ascend which contains the symlink of descend/and --> here/up. So that makes sense, it is following a symlink which it didn't before and finding a dir, "up" in the linked dir. Good. Looking at 1st difference... It is now showing up "lnk" which is a symlink: lnk --> dir So why is it showing this up and yet it is not showing descend/and (which is a link to here/up)? So yes we are following symlinks but are we supposed to just do the symlinks themselves as well? BTW, do we not allow user EAs on symlinks? (I've forgotten) --Tim