[PATCH v2 RESEND] xfs: fix dead loop at xfs_vm_writepage() on 32bit machine

[Jeff Liu]

On 09/26/2013 05:32 AM, Dave Chinner wrote:

> On Wed, Sep 25, 2013 at 04:10:20PM +0800, Jeff Liu wrote:
>> From: Jie Liu <jeff.liu at oracle.com>
>>
>> Write a file with an offset greater than 16TB on 32-bit system and
>> then trigger page write-back via sync(1) as below will cause the
>> task hang in a little while:
> 
> [snip]
> 
>> This patch just fixed both issues.
>>
>> Reported-by: Michael L. Semon <mlsemon35 at gmail.com>
>> Signed-off-by: Jie Liu <jeff.liu at oracle.com>
>> ---
>> v2: don't reset the s_max_bytes to MAX_LFS_FILESIZE, instead, revise the page offset
>>     check up strategy to avoid the potential overflow.
>> v1: http://oss.sgi.com/archives/xfs/2013-07/msg00154.html
>>
>>  fs/xfs/xfs_aops.c |   11 ++++++-----
>>  1 file changed, 6 insertions(+), 5 deletions(-)
>>
>> diff --git a/fs/xfs/xfs_aops.c b/fs/xfs/xfs_aops.c
>> index 41a6950..6059d00 100644
>> --- a/fs/xfs/xfs_aops.c
>> +++ b/fs/xfs/xfs_aops.c
>> @@ -969,7 +969,9 @@ xfs_vm_writepage(
>>  	offset = i_size_read(inode);
>>  	end_index = offset >> PAGE_CACHE_SHIFT;
>>  	last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
>> -	if (page->index >= end_index) {
>> +	if (page->index < end_index)
>> +		end_offset = (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT;
>> +	else {
>>  		unsigned offset_into_page = offset & (PAGE_CACHE_SIZE - 1);
> 
> The logic here is already difficult to understand, and the fact that
> the code that has 32 bit overflow issues is not obvious . Can you
> add a comment noting the overflow issue being handled here?
> 
>>  
>>  		/*
>> @@ -978,7 +980,8 @@ xfs_vm_writepage(
>>  		 * page so that reclaim stops reclaiming it. Otherwise
>>  		 * xfs_vm_releasepage() is called on it and gets confused.
>>  		 */
>> -		if (page->index >= end_index + 1 || offset_into_page == 0)
>> +		if (page->index > end_index ||
>> +		    (page->index == end_index && offset_into_page == 0))
>>  			goto redirty;
> 
> And again here? 
> 
> That means in future we will be aware of the problem when reading
> the code...

Fair enough, will post a new version at a latter time.

Thanks,
-Jeff