Date:	Thu, 17 Oct 2013 23:48:24 +0800
From:	Weijie Yang <weijie.yang.kh@...il.com>
To:	Hugh Dickins <hughd@...gle.com>
Cc:	Krzysztof Kozlowski <k.kozlowski@...sung.com>,
	Al Viro <viro@...iv.linux.org.uk>,
	Andrew Morton <akpm@...ux-foundation.org>,
	Linux-Kernel <linux-kernel@...r.kernel.org>,
	Linux-MM <linux-mm@...ck.org>, Michal Hocko <mhocko@...e.cz>,
	Shaohua Li <shli@...ionio.com>,
	Minchan Kim <minchan@...nel.org>
Subject: Re: [PATCH] swap: fix setting PAGE_SIZE blocksize during
 swapoff/swapon race

On Wed, Oct 16, 2013 at 1:19 AM, Hugh Dickins <hughd@...gle.com> wrote:
> On Tue, 15 Oct 2013, Krzysztof Kozlowski wrote:
>> On Tue, 2013-10-15 at 02:59 -0700, Hugh Dickins wrote:
>> > On Mon, 14 Oct 2013, Krzysztof Kozlowski wrote:
>> >
>> > > Fix race between swapoff and swapon resulting in setting blocksize of
>> > > PAGE_SIZE for block devices during swapoff.
>> > >
>> > > The swapon modifies swap_info->old_block_size before acquiring
>> > > swapon_mutex. It reads block_size of bdev, stores it under
>> > > swap_info->old_block_size and sets new block_size to PAGE_SIZE.
>> > >
>> > > On the other hand the swapoff sets the device's block_size to
>> > > old_block_size after releasing swapon_mutex.
>> > >
>> > > This patch locks the swapon_mutex much earlier during swapon. It also
>> > > releases the swapon_mutex later during swapoff.
>> > >
>> > > The effect of race can be triggered by following scenario:
>> > >  - One block swap device with block size of 512
>> > >  - thread 1: Swapon is called, swap is activated,
>> > >    p->old_block_size = block_size(p->bdev); /512/
>> > >    block_size(p->bdev) = PAGE_SIZE;
>> > >    Thread ends.
>> > >
>> > >  - thread 2: Swapoff is called and it goes just after releasing the
>> > >    swapon_mutex. The swap is now fully disabled except of setting the
>> > >    block size to old value. The p->bdev->block_size is still equal to
>> > >    PAGE_SIZE.
>> > >
>> > >  - thread 3: New swapon is called. This swap is disabled so without
>> > >    acquiring the swapon_mutex:
>> > >    - p->old_block_size = block_size(p->bdev); /PAGE_SIZE (!!!)/
>> > >    - block_size(p->bdev) = PAGE_SIZE;
>> > >    Swap is activated and thread ends.
>> > >
>> > >  - thread 2: resumes work and sets blocksize to old value:
>> > >    - set_blocksize(bdev, p->old_block_size)
>> > >    But now the p->old_block_size is equal to PAGE_SIZE.
>> > >
>> > > The patch swap-fix-set_blocksize-race-during-swapon-swapoff does not fix
>> > > this particular issue. It reduces the possibility of races as the swapon
>> > > must overwrite p->old_block_size before acquiring swapon_mutex in
>> > > swapoff.
>> > >
>> > > Signed-off-by: Krzysztof Kozlowski <k.kozlowski@...sung.com>
>> >
>> > Sorry you're being blown back and forth on this, but I say Nack to
>> > this version.  I've not spent the time to check whether it ends up
>> > correct or not; but your original patch was appropriate to the bug,
>> > and this one is just unnecessary churn in my view.
>>
>> Hi,
>>
>> I still think my previous patch does not solve the issue entirely.
>> The call set_blocksize() in swapoff quite often sets PAGE_SIZE instead
>> of valid block size (e.g. 512). I trigger this with:
>
> PAGE_SIZE and 512 are equally valid block sizes,
> it's just hard to support both consistently at the same instant.
>
>> ------
>> for i in `seq 1000`
>> do
>>       swapoff /dev/sdc1 &
>>       swapon /dev/sdc1 &
>>       swapon /dev/sdc1 &
>> done
>> ------
>> 10 seconds run of this script resulted in 50% of set_blocksize(PAGE_SIZE).
>> Although effect can only be observed after adding printks (block device is
>> released).
>
> But despite PAGE_SIZE being a valid block size,
> I agree that it's odd if you see variation there.
>
> Here's my guess: it looks as if the p->bdev test is inadequate, in the
> decision whether bad_swap should set_blocksize() or not: p->bdev is not
> usually reset when a swap_info_struct is released for reuse.
>
> Please try correcting that, either by resetting p->bdev where necessary,
> or by putting a better test in bad_swap: see if that fixes this oddity.
>
> I still much prefer your original little patch,
> to this extension of the use of swapon_mutex.
>
> However, a bigger question would be, why does swapoff have to set block
> size back to old_block_size anyway?  That was introduced in 2.5.13 by
>
> <viro@...h.psu.edu> (02/05/01 1.447.69.1)
>         [PATCH] (1/6) blksize_size[] removal
>
>          - preliminary cleanups: make sure that swapoff restores original block
>            size, kill set_blocksize() (and use of __bread()) in multipath.c,
>            reorder opening device and finding its block size in mtdblock.c.
>
> Al, not an urgent question, but is this swapoff old_block_size stuff
> still necessary?  And can't swapon just use whatever bd_block_size is
> already in force?  IIUC, it plays no part beyond the initial readpage
> of swap header.
>
> Thanks,
> Hugh

Let me try to explain(and guess):
we have to set_block in swapon. the swap_header is PAGE_SIZE, if device's
blocksize is more than PAGE_SIZE, then the swap entry address on swapfile
would be not PAGE_SIZE aligned. or one swap page can not fill a block.
There maybe a problem for some device.
The set_blocksize() do the judgement work for swapon.
And may be some userland tools assume swap device blocksize is PAGE_SIZE？

issues here are more than this one:
After swap_info_struct is released for reuse in swapoff.
Its corresponding resources are released later, such as:
- swap_cgroup_swapoff(type);
- blkdev_put
- inode->i_flags &= ~S_SWAPFILE;

we need release(or clean) these resources before release swap_info_struct.

to Krzysztof: I think it is better to add this handle to your patch

regards
--
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