Message-ID: <4ED6FA2C.7010005@oracle.com>
Date: Wed, 30 Nov 2011 21:53:16 -0600
From: Dave Kleikamp <dave.kleikamp@oracle.com>
MIME-Version: 1.0
To: Chris Dunlop <chris@onthe.net.au>
CC: David Howells <dhowells@redhat.com>, Al Viro <viro@zeniv.linux.org.uk>,
        "Myklebust, Trond" <Trond.Myklebust@netapp.com>,
        linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org,
        Eric Van Hensbergen <ericvh@gmail.com>,
        Ron Minnich <rminnich@sandia.gov>, Latchesar Ionkov <lucho@ionkov.net>,
        Jan Harkes <jaharkes@cs.cmu.edu>,
        "\"maintainer@onthe.net.au\":CODA FILE SYSTEM" <coda@cs.cmu.edu>,
        Petr Vandrovec <petr@vandrovec.name>,
        Greg Kroah-Hartman <gregkh@suse.de>,
        v9fs-developer@lists.sourceforge.net, linux-afs@lists.infradead.org,
        codalist@TELEMANN.coda.cs.cmu.edu,
        jfs-discussion@lists.sourceforge.net, linux-nfs@vger.kernel.org
Subject: Re: [PATCH 1/1] fix d_revalidate oopsen on NFS exports
References: <20111130071319.GA16711@onthe.net.au> <1321861008-20611-1-git-send-email-chris@onthe.net.au> <20111129082501.GA569@onthe.net.au> <2E1EB2CF9ED1CB4AA966F0EB76EAB4430C3CBC20@SACMVEXC2-PRD.hq.netapp.com> <24960.1322643283@redhat.com> <20111201004709.GA26085@onthe.net.au> <4ED6E4EF.7050101@oracle.com> <20111201033346.GA1049@onthe.net.au>
In-Reply-To: <20111201033346.GA1049@onthe.net.au>
Content-Type: text/plain; charset=ISO-8859-1
Sender: linux-nfs-owner@vger.kernel.org


On 11/30/2011 09:33 PM, Chris Dunlop wrote:
> On Wed, Nov 30, 2011 at 08:22:39PM -0600, Dave Kleikamp wrote:
>> On 11/30/2011 06:47 PM, Chris Dunlop wrote:
>>>> It's also worth printing a message - this *is* a kernel bug of some description
>>>> if it happens.
>>>
>>> Like the below?  This covers the d_revalidate for 9p, afs, coda,
>>> hfs, ncpfs, proc, sysfs.
>>>
>>> Note:  jfs isn't susceptible to this problem, but the resolution
>>> doesn't look like the other file systems, and from the comment
>>> I'm not sure if the problem was really understood and if it's
>>> doing the right thing:
>>
>> This code, as well as the comments, were copied from vfat. It seems
>> reasonable for case-insensitive but case-preserving behavior (not jfs's
>> default). The safe thing is to drop the negative dentry if we don't know
>> the operation.
> 
> In that case, it looks like the thing to do might be to add the
> "protection" to the start of jfs_ci_revaliate(), per how the
> original has been changed in vfat:

The LOOKUP_RCU check had previously been there, but Al Viro removed it:

commit 5c0f360b083fb33d05d1bff4b138b82d715eb419
Author: Al Viro <viro@zeniv.linux.org.uk>
Date:   Sat Jun 25 21:41:09 2011 -0400

    jfs_ci_revalidate() is safe from RCU mode

I'm not sure what it takes to be "safe", but this is a simple function
that doesn't block, take locks, or do much of anything. You shouldn't
need to do anything with jfs.

Shaggy

> 
> fs/fat/namei_vfat.c:
> static int vfat_revalidate_ci(struct dentry *dentry, struct nameidata *nd)
> {
>         if (nd && nd->flags & LOOKUP_RCU)
>                 return -ECHILD;
>         ...
> }
> 
> E.g.:
> 
> ----------------------------------------------------------------------
> Don't oops when abused by broken layered file systems
> 
> Signed-off-by: Chris Dunlop <chris@onthe.net.au>
> ---
>  fs/jfs/namei.c |    3 +++
>  1 files changed, 3 insertions(+), 0 deletions(-)
> 
> diff --git a/fs/jfs/namei.c b/fs/jfs/namei.c
> index e17545e..5504f6e 100644
> --- a/fs/jfs/namei.c
> +++ b/fs/jfs/namei.c
> @@ -1585,6 +1585,9 @@ out:
>  
>  static int jfs_ci_revalidate(struct dentry *dentry, struct nameidata *nd)
>  {
> +	if (nd && nd->flags & LOOKUP_RCU)
> +		return -ECHILD;
> +
>  	/*
>  	 * This is not negative dentry. Always valid.
>  	 *