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] [thread-next>] [day] [month] [year] [list]
Date:   Fri, 27 Apr 2018 10:37:27 +0800
From:   Chao Yu <yuchao0@...wei.com>
To:     Jaegeuk Kim <jaegeuk@...nel.org>, Chao Yu <chao@...nel.org>
CC:     <linux-f2fs-devel@...ts.sourceforge.net>,
        <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] f2fs: fix to wait IO writeback in __revoke_inmem_pages()

On 2018/4/27 0:36, Jaegeuk Kim wrote:
> On 04/26, Chao Yu wrote:
>> On 2018/4/26 23:48, Jaegeuk Kim wrote:
>>> On 04/26, Chao Yu wrote:
>>>> Thread A				Thread B
>>>> - f2fs_ioc_commit_atomic_write
>>>>  - commit_inmem_pages
>>>>   - f2fs_submit_merged_write_cond
>>>>   : write data
>>>> 					- write_checkpoint
>>>> 					 - do_checkpoint
>>>> 					 : commit all node within CP
>>>> 					 -> SPO
>>>>   - f2fs_do_sync_file
>>>>    - file_write_and_wait_range
>>>>    : wait data writeback
>>>>
>>>> In above race condition, data/node can be flushed in reversed order when
>>>> coming a checkpoint before f2fs_do_sync_file, after SPOR, it results in
>>>> atomic written data being corrupted.
>>>
>>> Wait, what is the problem here? Thread B could succeed checkpoint, there is
>>> no problem. If it fails, there is no fsync mark where we can recover it, so
>>
>> Node is flushed by checkpoint before data, with reversed order, that's the problem.
> 
> What do you mean? Data should be in disk, in order to proceed checkpoint.

1. thread A: commit_inmem_pages submit data into block layer, but haven't waited
it writeback.
2. thread A: commit_inmem_pages update related node.
3. thread B: do checkpoint, flush all nodes to disk
4. SPOR

Then, atomic file becomes corrupted since nodes is flushed before data.

Thanks,

> 
>>
>> Thanks,
>>
>>> we can just ignore the last written data as nothing.
>>>
>>>>
>>>> This patch adds f2fs_wait_on_page_writeback in __revoke_inmem_pages() to
>>>> keep data and node of atomic file being flushed orderly.
>>>>
>>>> Signed-off-by: Chao Yu <yuchao0@...wei.com>
>>>> ---
>>>>  fs/f2fs/file.c    | 4 ++++
>>>>  fs/f2fs/segment.c | 3 +++
>>>>  2 files changed, 7 insertions(+)
>>>>
>>>> diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c
>>>> index be7578774a47..a352804af244 100644
>>>> --- a/fs/f2fs/file.c
>>>> +++ b/fs/f2fs/file.c
>>>> @@ -217,6 +217,9 @@ static int f2fs_do_sync_file(struct file *file, loff_t start, loff_t end,
>>>>  
>>>>  	trace_f2fs_sync_file_enter(inode);
>>>>  
>>>> +	if (atomic)
>>>> +		goto write_done;
>>>> +
>>>>  	/* if fdatasync is triggered, let's do in-place-update */
>>>>  	if (datasync || get_dirty_pages(inode) <= SM_I(sbi)->min_fsync_blocks)
>>>>  		set_inode_flag(inode, FI_NEED_IPU);
>>>> @@ -228,6 +231,7 @@ static int f2fs_do_sync_file(struct file *file, loff_t start, loff_t end,
>>>>  		return ret;
>>>>  	}
>>>>  
>>>> +write_done:
>>>>  	/* if the inode is dirty, let's recover all the time */
>>>>  	if (!f2fs_skip_inode_update(inode, datasync)) {
>>>>  		f2fs_write_inode(inode, NULL);
>>>> diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
>>>> index 584483426584..9ca3d0a43d93 100644
>>>> --- a/fs/f2fs/segment.c
>>>> +++ b/fs/f2fs/segment.c
>>>> @@ -230,6 +230,8 @@ static int __revoke_inmem_pages(struct inode *inode,
>>>>  
>>>>  		lock_page(page);
>>>>  
>>>> +		f2fs_wait_on_page_writeback(page, DATA, true);
>>>> +
>>>>  		if (recover) {
>>>>  			struct dnode_of_data dn;
>>>>  			struct node_info ni;
>>>> @@ -415,6 +417,7 @@ static int __commit_inmem_pages(struct inode *inode)
>>>>  		/* drop all uncommitted pages */
>>>>  		__revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
>>>>  	} else {
>>>> +		/* wait all committed IOs writeback and release them from list */
>>>>  		__revoke_inmem_pages(inode, &revoke_list, false, false);
>>>>  	}
>>>>  
>>>> -- 
>>>> 2.15.0.55.gc2ece9dc4de6
> 
> .
> 

Powered by blists - more mailing lists