xfs
[Top] [All Lists]

Re: inode_permission NULL pointer dereference in 3.13-rc1

To: Dave Chinner <david@xxxxxxxxxxxxx>
Subject: Re: inode_permission NULL pointer dereference in 3.13-rc1
From: Al Viro <viro@xxxxxxxxxxxxxxxxxx>
Date: Thu, 28 Nov 2013 23:44:41 +0000
Cc: Christoph Hellwig <hch@xxxxxxxxxxxxx>, linux-fsdevel@xxxxxxxxxxxxxxx, Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx>, xfs@xxxxxxxxxxx
Delivered-to: xfs@xxxxxxxxxxx
In-reply-to: <20131128225102.GS10988@dastard>
References: <20131124140413.GA19271@xxxxxxxxxxxxx> <20131124152758.GL10323@xxxxxxxxxxxxxxxxxx> <20131125160648.GA4933@xxxxxxxxxxxxx> <20131126131134.GM10323@xxxxxxxxxxxxxxxxxx> <20131126141253.GA28062@xxxxxxxxxxxxx> <20131127064351.GN10323@xxxxxxxxxxxxxxxxxx> <20131127100906.GA19740@xxxxxxxxxxxxx> <20131128162618.GO10323@xxxxxxxxxxxxxxxxxx> <20131128212301.GP10323@xxxxxxxxxxxxxxxxxx> <20131128225102.GS10988@dastard>
Sender: Al Viro <viro@xxxxxxxxxxxxxxxx>
User-agent: Mutt/1.5.21 (2010-09-15)
On Fri, Nov 29, 2013 at 09:51:02AM +1100, Dave Chinner wrote:

> > Looks like adding if (!nd->inode) { a bunch of printks } in the end of
> > path_init() makes the sucker disappear (so far 2 times out of 2, and
> > with a test run taking a bit under two hours, well...)  The plain
> > WARN_ON(!nd->inode) in that place triggers just fine.
> 
> I usually find that when printk() makes race conditions go away,
> switching to tracepoints works better. It's still not as good as
> reliable as when the debug is not there, but it seems to perturb
> race conditions a lot less.

Actually, I've just got the output from this run, and it's really interesting.
We get path_init() setting NULL nd->inode for open() of "/dev/ptmx" (from
/sbin/startpar).  And what we have at the time we get to link_path_walk() is
        * LOOKUP_RCU | LOOKUP_FOLLOW | LOOKUP_PARENT | LOOKUP_JUMPED in
nd->flags (as expected)
        * current->fs->root, current->fs->pwd and nd->path being the same
vfsmount/dentry pair.
        * dentry in question has ->d_sb->s_id containing "sda1", as expected
for root fs.
        * ->mnt_root of that vfsmount being equal to dentry
So far, so good, right?
        * d_count(dentry) is -128
        * dentry->d_inode is NULL

In other words, what we get is an extra dput() somewhere.  The trouble is,
all likely places I'm seeing in the "RCU'd vfsmounts" seem to be OK...
In theory, we might be hitting a _missing_ dput(), with counter wrapping
around, but that doesn't seem likely...

<Prev in Thread] Current Thread [Next in Thread>