Re: [RFC PATCH] mark buffer_head mapping preallocate area as new during write_begin with delayed allocation

[Eric Sandeen <sandeen@redhat.com>]

Jan Kara wrote:
>> Aneesh Kumar K.V wrote:
>>> On Tue, Apr 28, 2009 at 01:00:47PM -0400, Theodore Tso wrote:
>>>> On Tue, Apr 28, 2009 at 10:05:54PM +0530, Aneesh Kumar K.V wrote:
>> ...
>>>>>> The other problem seems to be in the case of a delayed allocation
>>>>>> write, where we return a buffer_head which is marked new, and this
>>>>>> causes block_prepare_write() to call unmap_underlying_metadata(dev, 0).
>>>>> Not just that. On block allocation we are not calling
>>>>> unmap_underlying_metadata(dev, blocknumber) for delayed allocated
>>>>> blocks. That would imply file corruption.
>>>> I don't think I'm following you .  If we write into block that was
>>>> delayed allocated.  Are you saying we might get in trouble of the
>>>> delayed allocation block is mmap'ed in?
>>> We allocate blocks for delayed buffer during writepage. Now we need to
>>> make sure after getting the blocks we drop the old buffer_head mapping
>>> that we may have with this particular block attached to the block
>>> device. That is done by calling unmap_underlying_metadata. Now the
>>> current code doesn't call unmap_underlying_metadata for delayed
>>> allocated blocks. That would mean we can see corrupt files if old
>>> buffer_head mapping gets synced to disk AFTER we write the new
>>> buffer_head mapping.
>>
>> Talking w/ Aneesh on IRC, I don't see how we can have stray dirty
>> mappings lying around for this block device unless someone is writing
>> directly to the mounted block device, which I don't think is ever
>> considered safe ...
>>
>> I'm not quite sure what the call to __unmap_underlying_blocks() in
>> mpage_da_map_blocks() is for, I guess?
>   For ext3 / ext4 I think we don't need unmap_underlying_blocks() since
> before we reallocate a block, we make sure that the transaction freeing
> the block is committed and clear all dirty bits from freed blocks.
>   But for more careless filesystems, if they reallocate metadata block
> as a data block and don't clear the dirty bit in blockdev mapping,
> unmap_underlying_blocks() does it for them.

That's what I thought - so I was wondering why we have specific calls to
this in ext4:

mpage_da_map_blocks
	__unmap_underlying_blocks
		for (i = 0; i < blocks; i++)
			unmap_underlying_metadata

-Eric