Re: [pnfs] [GIT BISECT] first bad commit: 1f36f774 Switch !O_CREAT case to use of do_last()

From: Benny Halevy <bhalevy@panasas.com>
To: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Boaz Harrosh <bharrosh@panasas.com>,
	linux-fsdevel <linux-fsdevel@vger.kernel.org>,
	"J. Bruce Fields" <bfields@citi.umich.edu>,
	pNFS Mailing List <pnfs@linux-nfs.org>,
	linux-kernel <linux-kernel@vger.kernel.org>,
	Doug Nazar <nazard.lkml@gmail.com>,
	Trond Myklebust <Trond.Myklebust@netapp.com>
Subject: Re: [pnfs] [GIT BISECT] first bad commit: 1f36f774 Switch !O_CREAT case to use of do_last()
Date: Thu, 25 Mar 2010 12:56:40 +0200	[thread overview]
Message-ID: <4BAB4168.2090404@panasas.com> (raw)
In-Reply-To: <20100325104924.GV30031@ZenIV.linux.org.uk>

On Mar. 25, 2010, 12:49 +0200, Al Viro <viro@ZenIV.linux.org.uk> wrote:
> On Thu, Mar 25, 2010 at 12:22:31PM +0200, Benny Halevy wrote:
>> On Mar. 25, 2010, 12:12 +0200, Al Viro <viro@ZenIV.linux.org.uk> wrote:
>>> On Thu, Mar 25, 2010 at 11:39:38AM +0200, Boaz Harrosh wrote:
>>>
>>>> It makes no difference, fails just the same. Would an "strace" help?
>>> It might, especially if you ran it for identical repositories on local
>>> fs and on NFS; at least that way it would be possible to see where do
>>> they diverge...
>>> _______________________________________________
>>> pNFS mailing list
>>> pNFS@linux-nfs.org
>>> http://linux-nfs.org/cgi-bin/mailman/listinfo/pnfs
>> ext3 fs exported on the server
>> client mount -t nfs4 localhost:/ /mnt/localhost
>> $ strace git status
>> ...
>> open(".git/objects/pack", O_RDONLY|O_NONBLOCK|O_DIRECTORY|0x80000) = -1 EISDIR (Is a directory)
> 
> Interesting.  It is a directory, indeed, but why the hell does that call
> fail with -EISDIR?
> 
> Does that happen with nfsv3 or is that v4-only?  I'm going to set up v4
> server and client and see what happens, but that information could be
> useful...

We've seen this with v4 only so far.

BTW I added this WARN_ON:
@@ -1656,8 +1659,10 @@ static struct file *do_last(struct nameidata *nd, struct path *path,
 		if (path->dentry->d_inode->i_op->follow_link)
 			return NULL;
 		error = -ENOTDIR;
-		if (*want_dir && !path->dentry->d_inode->i_op->lookup)
+		if (*want_dir && !path->dentry->d_inode->i_op->lookup) {
+			WARN_ON(1);
 			goto exit_dput;
+		}
 		path_to_nameidata(path, nd);
 		audit_inode(pathname, nd->path.dentry);
 		goto ok;

but it is NOT tripping for this scenario.

(for some reason I saw it tripping when building the kernel over nfs
but it's benign)

Benny
