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  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date:	Tue, 6 Oct 2015 16:12:18 +1100
From:	Dave Chinner <david@...morbit.com>
To:	Andrew Morton <akpm@...ux-foundation.org>
Cc:	Michal Hocko <mhocko@...nel.org>,
	Guenter Roeck <linux@...ck-us.net>,
	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
	Christoph Lameter <cl@...ux.com>,
	"Kirill A. Shutemov" <kirill.shutemov@...ux.intel.com>,
	linux-mm@...ck.org
Subject: Re: linux-next: kernel BUG at mm/slub.c:1447!

On Mon, Oct 05, 2015 at 07:12:17PM -0700, Andrew Morton wrote:
> On Mon, 5 Oct 2015 12:29:36 -0700 Andrew Morton <akpm@...ux-foundation.org> wrote:
> 
> > Maybe it would be better to add the gfp_t argument to the
> > address_space_operations.  At a minimum, writepage(), readpage(),
> > writepages(), readpages().  What a pickle.
> 
> I'm being dumb.  All we need to do is to add a new
> 
> 	address_space_operations.readpage_gfp(..., gfp_t gfp)
> 
> etc.  That should be trivial.  Each fs type only has 2-4 instances of
> address_space_operations so the overhead is miniscule.
> 
> As a background janitorial thing we can migrate filesystems over to the
> new interface then remove address_space_operations.readpage() etc.  But
> this *would* add overhead: more stack and more text for no gain.  So
> perhaps we just leave things as they are.
> 
> That's so simple that I expect we can short-term fix this bug with that
> approach.  umm, something like
> 
> --- a/fs/mpage.c~a
> +++ a/fs/mpage.c
> @@ -139,7 +139,8 @@ map_buffer_to_page(struct page *page, st
>  static struct bio *
>  do_mpage_readpage(struct bio *bio, struct page *page, unsigned nr_pages,
>  		sector_t *last_block_in_bio, struct buffer_head *map_bh,
> -		unsigned long *first_logical_block, get_block_t get_block)
> +		unsigned long *first_logical_block, get_block_t get_block,
> +		gfp_t gfp)

Simple enough, but not sufficient to avoid deadlocks because this
doesn't address the common vector for deadlock that was reported.
i.e. deadlocks due to the layering dependency the loop device
creates between two otherwise independent filesystems.

If read IO through the loop device does GFP_KERNEL allocations, then
it is susceptible to deadlock as that can force writeback and
transactions from the filesystem on top of the loop device, which
does more IO to the loop device, which then backs up on the backing
file inode mutex. Forwards progress cannot be made until the
GFP_KERNEL allocation succeeds, but that depends on the loop device
making forwards progress...

The loop device indicates this is a known problems tries to avoid
these deadlocks by doing this on it's backing file:

	mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS))

to try to ensure that mapping related allocations do not cause
inappropriate reclaim contexts to be triggered.

This is a problem independent of any specific filesystem - let's not
hack a workaround into a specific filesystem just because the
problem was reported on that filesystem....

Cheers,

Dave.
-- 
Dave Chinner
david@...morbit.com
--
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/

Powered by blists - more mailing lists