linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Mimi Zohar <zohar@linux.vnet.ibm.com>
To: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Krisztian Litkey <kli@iki.fi>,
	linux-unionfs@vger.kernel.org,
	Krisztian Litkey <krisztian.litkey@intel.com>,
	linux-security-module <linux-security-module@vger.kernel.org>,
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH v3 1/1] ovl: setxattr: don't deadlock when called from ima_fix_xattr.
Date: Fri, 20 May 2016 13:00:33 -0400	[thread overview]
Message-ID: <1463763633.2763.34.camel@linux.vnet.ibm.com> (raw)
In-Reply-To: <20160520162937.GU14480@ZenIV.linux.org.uk>

On Fri, 2016-05-20 at 17:29 +0100, Al Viro wrote:
> On Fri, May 20, 2016 at 10:21:27AM -0400, Mimi Zohar wrote:
> > > +	if (mutex_is_locked(&upper->d_inode->i_mutex))
> > > +		err = __vfs_setxattr_noperm(upper, name, value, size, flags);
> > 
> > As far as I'm aware, the only time that i_mutex is taken, is during
> > __fput() when IMA writes security.ima.   Previous versions of this patch
> > checked whether the xattr being written was security.ima.  It would
> > probably be a good idea not to make that assumption here.   The question
> > is what should happen if the i_mutex is locked, but the xattr isn't
> > security.ima.  At minimum it should be audited.  Al, any comments?
> 
> ITYM "printable", and that's somewhat harder.  OK, let me try:
> 
> Anybody using ..._is_lock() kind of primitives that way ought to be
> (re)educated before they are allowed near any kind of multithreaded
> code _anywhere_.  mutex could've been held by a different thread of
> execution and dropped just as mutex_is_locked() returns.  Or at any
> subsequent point.  This is 100% bogus; one should *never* write that kind
> of code.  As in "here's your well-earned F-, better luck next semester".
> 
> I haven't seen the full patch (you've quoted only a part of that gem), but
> about the only way for it to be correct is to have it continue with
> + else
> +       err = <identical call>
> 
> 	Practically all calls of mutex_is_locked() are of form
> WARN_ON(!mutex_is_locked(...)) or equivalent thereof.  And the rest contains
> similar... wonders - for example, take a look at drivers/media/rc/imon.c;
> imon_ir_change_protocol() has this
>         if (!mutex_is_locked(&ictx->lock)) {
>                 unlock = true;
>                 mutex_lock(&ictx->lock);
>         }
> 
>         retval = send_packet(ictx);
>         if (retval)
>                 goto out;
> 
>         ictx->rc_type = *rc_type;
>         ictx->pad_mouse = false;
> 
> out:
>         if (unlock)
>                 mutex_unlock(&ictx->lock);
> Finding why it's exploitably racy is left as a trivial exercise for readers...
> 
> Folks, if you see something of that sort in the code, it's a huge red flag.
> There are legitimate uses of mutex_is_locked other than asserts, but those
> are extremely rare.

My fault for even suggesting it.

> I would need to see more context to be able to comment on the problem in
> question, but this patch is almost certainly broken.

We deferred __fput() back in 2012 in order for IMA to safely take the
i_mutex and write security.ima.   Writing the security.ima xattr now
triggers overlayfs to write the xattr, but overlayfs doesn't
differentiate between callers - as a result of userspace or as described
here in __fput().   All calls to ovl_setxattr() should call vfs_sexattr,
except the one triggered by __fput().   Refer to the original lockdep
report -    
http://thread.gmane.org/gmane.linux.file-systems.union/640

Al, any help in resolving this lockdep would be much appreciated.

Mimi

  reply	other threads:[~2016-05-20 17:01 UTC|newest]

Thread overview: 7+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
     [not found] <1463611500.2465.22.camel@linux.vnet.ibm.com>
     [not found] ` <1463725718-2461-1-git-send-email-kli@iki.fi>
     [not found]   ` <1463754087.2763.17.camel@linux.vnet.ibm.com>
2016-05-20 16:29     ` [PATCH v3 1/1] ovl: setxattr: don't deadlock when called from ima_fix_xattr Al Viro
2016-05-20 17:00       ` Mimi Zohar [this message]
2016-05-20 20:53         ` Krisztian Litkey
2016-05-30 14:10           ` Miklos Szeredi
2016-05-30 16:50             ` Al Viro
2016-05-31  2:15               ` Mimi Zohar
2016-05-31  2:29             ` Mimi Zohar

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=1463763633.2763.34.camel@linux.vnet.ibm.com \
    --to=zohar@linux.vnet.ibm.com \
    --cc=kli@iki.fi \
    --cc=krisztian.litkey@intel.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-security-module@vger.kernel.org \
    --cc=linux-unionfs@vger.kernel.org \
    --cc=viro@ZenIV.linux.org.uk \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).