linux-kernel - Re: kernel BUG at /build/buildd/linux-2.6.35/mm/filemap.c:128!

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <alpine.LSU.2.00.1012291231540.22566@sister.anvils>
Date:	Wed, 29 Dec 2010 12:54:21 -0800 (PST)
From:	Hugh Dickins <hughd@...gle.com>
To:	robert@...ecki.net
cc:	Miklos Szeredi <miklos@...redi.hu>,
	Andrew Morton <akpm@...ux-foundation.org>,
	Nick Piggin <npiggin@...nel.dk>, linux-kernel@...r.kernel.org,
	linux-mm@...ck.org
Subject: Re: kernel BUG at /build/buildd/linux-2.6.35/mm/filemap.c:128!

On Tue, 30 Nov 2010, Hugh Dickins wrote:
> On Mon, 29 Nov 2010, Andrew Morton wrote:
> > On Tue, 23 Nov 2010 15:55:31 +0100
> > Robert  wi cki <robert@...ecki.net> wrote:
> > > >> [25142.286531] kernel BUG at /build/buildd/linux-2.6.35/mm/filemap.c:128!
> > > >
> > > > That's
> > > >
> > > >        BUG_ON(page_mapped(page));
> > > >
> > > > in  remove_from_page_cache().  That state is worth a BUG().
> > 
> > At a guess I'd say that another thread came in and established a
> > mapping against a page in the to-be-truncated range while
> > vmtruncate_range() was working on it.  In fact I'd be suspecting that
> > the mapping was established after truncate_inode_page() ran its
> > page_mapped() test.
> 
> It looks that way, but I don't see how it can be: the page is locked
> before calling truncate_inode_page() and unlocked after it: and the
> page (certainly in this tmpfs case, perhaps not for every filesystem)
> cannot be faulted into an address space without holding its page lock.
> 
> Either we've made a change somewhere, and are now dropping and retaking
> page lock in a way which exposes this bug?  Or truncate_inode_page()'s
> unmap_mapping_range() call is somehow missing the page it's called for?
> 
> I guess the latter is the more likely: maybe the truncate_count/restart
> logic isn't working properly.  I'll try to check over that again later -
> but will be happy if someone else beats me to it.

I have since found an omission in the restart_addr logic: looking back
at the October 2004 history of vm_truncate_count, I see that originally
I designed it to work one way, but hurriedly added a 7/6 redesign when
vma splitting turned out to leave an ambiguity.  I should have updated
the protection in mremap move at that time, but missed it.

Robert, please try out the patch below (should apply fine to 2.6.35):
I'm hoping this will fix what the fuzzer found, but it's still quite
possible that it found something else wrong that I've not yet noticed.
The patch could probably be cleverer (if we exported the notion of
restart_addr out of mm/memory.c), but I'm more in the mood for being
safe than clever at the moment.

I didn't hear whether you'd managed to try out Miklos's patch; but
this one is a better bet to be the fix for your particular issue.

Thanks,
Hugh

--- 2.6.37-rc8/mm/mremap.c	2010-11-01 13:01:32.000000000 -0700
+++ linux/mm/mremap.c	2010-12-29 12:25:46.000000000 -0800
@@ -91,9 +91,7 @@ static void move_ptes(struct vm_area_str
 		 */
 		mapping = vma->vm_file->f_mapping;
 		spin_lock(&mapping->i_mmap_lock);
-		if (new_vma->vm_truncate_count &&
-		    new_vma->vm_truncate_count != vma->vm_truncate_count)
-			new_vma->vm_truncate_count = 0;
+		new_vma->vm_truncate_count = 0;
 	}
 
 	/*
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/