From mboxrd@z Thu Jan  1 00:00:00 1970
From: sougata santra <sougata@tuxera.com>
Subject: Re: [PATCH] hfsplus: fix concurrent acess of alloc_blocks
Date: Wed, 19 Feb 2014 11:53:18 +0200
Message-ID: <53047F0E.7040705@tuxera.com>
References: <1392639647.29849.11.camel@ultrabook> <20140218140642.fdc6a406f22de69bdf73e66e@linux-foundation.org>
Mime-Version: 1.0
Content-Type: text/plain; charset="ISO-8859-1"; format=flowed
Content-Transfer-Encoding: 7bit
Cc: Christoph Hellwig <hch@lst.de>, <linux-fsdevel@vger.kernel.org>,
	<linux-kernel@vger.kernel.org>,
	Vyacheslav Dubeyko <slava@dubeyko.com>,
	Joe Perches <joe@perches.com>,
	Alexey Khoroshilov <khoroshilov@ispras.ru>
To: Andrew Morton <akpm@linux-foundation.org>
Return-path: <linux-fsdevel-owner@vger.kernel.org>
Received: from nbl-ex10-fe01.nebula.fi ([188.117.32.121]:57262 "EHLO
	ex10.nebula.fi" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org
	with ESMTP id S1753229AbaBSJxV (ORCPT
	<rfc822;linux-fsdevel@vger.kernel.org>);
	Wed, 19 Feb 2014 04:53:21 -0500
In-Reply-To: <20140218140642.fdc6a406f22de69bdf73e66e@linux-foundation.org>
Sender: linux-fsdevel-owner@vger.kernel.org
List-ID: <linux-fsdevel.vger.kernel.org>

On 02/19/2014 12:06 AM, Andrew Morton wrote:
> On Mon, 17 Feb 2014 14:20:47 +0200 Sougata Santra <sougata@tuxera.com> wrote:
>
>>
>> Concurrent access to alloc_blocks in hfsplus_inode_info is
>> protected by extents_lock mutex. This patch fixes two
>> instances where alloc_blocks modification was not protected
>> with this lock. This fixes possible allocation bitmap
>> corruption in race conditions while extending and truncating
>> files.
>>
>> ...
>>
>> --- a/fs/hfsplus/extents.c
>> +++ b/fs/hfsplus/extents.c
>> @@ -498,11 +498,13 @@ int hfsplus_file_extend(struct inode *inode)
>>   			goto insert_extent;
>>   	}
>>   out:
>> -	mutex_unlock(&hip->extents_lock);
>>   	if (!res) {
>>   		hip->alloc_blocks += len;
>> +		mutex_unlock(&hip->extents_lock);
>>   		hfsplus_mark_inode_dirty(inode, HFSPLUS_I_ALLOC_DIRTY);
>> +		return 0;
>>   	}
>> +	mutex_unlock(&hip->extents_lock);
>>   	return res;
>>
>
> This looks OK.
>
>> @@ -592,9 +594,9 @@ void hfsplus_file_truncate(struct inode *inode)
>>   		hfs_brec_remove(&fd);
>>   	}
>>   	hfs_find_exit(&fd);
>> -	mutex_unlock(&hip->extents_lock);
>>
>>   	hip->alloc_blocks = blk_cnt;
>> +	mutex_unlock(&hip->extents_lock);
>>   out:
>>   	hip->phys_size = inode->i_size;
>>   	hip->fs_blocks = (inode->i_size + sb->s_blocksize - 1) >>
>
> But this does not.  To provide locking for
> hfsplus_inode_info.alloc_blocks, we must take the lock *before* taking
> a local copy of ->alloc_blocks.
>
> Please review:
>
> --- a/fs/hfsplus/extents.c~hfsplus-fix-concurrent-acess-of-alloc_blocks-fix
> +++ a/fs/hfsplus/extents.c
> @@ -556,11 +556,13 @@ void hfsplus_file_truncate(struct inode
>
>   	blk_cnt = (inode->i_size + HFSPLUS_SB(sb)->alloc_blksz - 1) >>
>   			HFSPLUS_SB(sb)->alloc_blksz_shift;
> +
> +	mutex_lock(&hip->extents_lock);
> +
>   	alloc_cnt = hip->alloc_blocks;
>   	if (blk_cnt == alloc_cnt)
> -		goto out;
> +		goto out_unlock;
>
> -	mutex_lock(&hip->extents_lock);
>   	res = hfs_find_init(HFSPLUS_SB(sb)->ext_tree, &fd);
>   	if (res) {
>   		mutex_unlock(&hip->extents_lock);
> @@ -594,6 +596,7 @@ void hfsplus_file_truncate(struct inode
>   	hfs_find_exit(&fd);
>
>   	hip->alloc_blocks = blk_cnt;
> +out_unlock:
>   	mutex_unlock(&hip->extents_lock);
>   out:
We can also remove this label ?
>   	hip->phys_size = inode->i_size;
> _
>