[Ocfs2-devel] [PATCH 02/15] ocfs2: Add clusters free in dealloc_ctxt.

[Tao Ma]

Joel Becker wrote:
> On Tue, Nov 04, 2008 at 09:34:40AM +0800, Tao Ma wrote:
>> Joel Becker wrote:
>>>> +{
>>>> +	struct ocfs2_cached_block_free *tmp;
>>>> +	struct inode *tl_inode = osb->osb_tl_inode;
>>>> +	handle_t *handle;
>>>> +	int ret = 0;
>>>> +
>>>> +	mutex_lock(&tl_inode->i_mutex);
>>>> +
>>>> +	handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
>>>> +	if (IS_ERR(handle)) {
>>>> +		ret = PTR_ERR(handle);
>>>> +		mlog_errno(ret);
>>>> +		goto out_mutex;
>>>> +	}
>>>> +
>>>> +	while (head) {
>>>> +		ret = ocfs2_truncate_log_append(osb, handle, head->free_blk,
>>>> +						head->free_bit);
>>>> +		if (ret) {
>>>> +			mlog_errno(ret);
>>>> +			break;
>>>> +		}
>>> 	What happens if the truncate log fills up?
>> currently, it is only used in one xattr set. So the total size is only  
>> 64K bytes and I don't think the chance is very high. I have also gone  
>> through ocfs2_commit_truncate, no error handler there, and even if it is  
>> filled up and error out, fsck should be able to find them. But you are  
>> probably right, if in future we use it in truncating file, it may be a  
>> very high risk. But it could be in another separate patch, agree?
> 
> 	ocfs2_commit_truncate() is a loop from start: to bail:.  It
> calls ocfs2_do_truncate() for each extent it removes.  Before each call
> to ocfs2_do_truncate(), it checks ocfs2_truncate_log_needs_flush().
> That is, it never adds more than one record to the truncate log at a
> time, and it checks for a full log before adding each record.
> 	Your 64K can be 16 clusters at a 4K cluster size.  That's 16
> records.  What if the truncate log had only space for three records from
> a previous operation?  Your fourth call to ocfs2_truncate_log_append()
> will mlog_bug_on_msg() like so:
> 
> 5319         mlog_bug_on_msg(tl_count > ocfs2_truncate_recs_per_inode(osb->sb) ||
> 
> 	That is not acceptable.  We don't introduce known bugs and say
> they are OK if they are rare.
Fair enough. I will try to find a way for it. thanks.

Regards,
Tao