Re: [PATCH v2 3/6] ext4: restart handle if credits are insufficient during allocating blocks

[Zhang Yi <yi.zhang@huaweicloud.com>]

On 2025/6/20 22:18, Jan Kara wrote:
> On Fri 20-06-25 13:00:32, Zhang Yi wrote:
>> On 2025/6/20 0:33, Jan Kara wrote:
>>> On Wed 11-06-25 19:16:22, Zhang Yi wrote:
>>>> From: Zhang Yi <yi.zhang@huawei.com>
>>>>
>>>> After large folios are supported on ext4, writing back a sufficiently
>>>> large and discontinuous folio may consume a significant number of
>>>> journal credits, placing considerable strain on the journal. For
>>>> example, in a 20GB filesystem with 1K block size and 1MB journal size,
>>>> writing back a 2MB folio could require thousands of credits in the
>>>> worst-case scenario (when each block is discontinuous and distributed
>>>> across different block groups), potentially exceeding the journal size.
>>>> This issue can also occur in ext4_write_begin() and ext4_page_mkwrite()
>>>> when delalloc is not enabled.
>>>>
>>>> Fix this by ensuring that there are sufficient journal credits before
>>>> allocating an extent in mpage_map_one_extent() and _ext4_get_block(). If
>>>> there are not enough credits, return -EAGAIN, exit the current mapping
>>>> loop, restart a new handle and a new transaction, and allocating blocks
>>>> on this folio again in the next iteration.
>>>>
>>>> Suggested-by: Jan Kara <jack@suse.cz>
>>>> Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
>>>
>>> ...
>>>
>>>>  static int _ext4_get_block(struct inode *inode, sector_t iblock,
>>>>  			   struct buffer_head *bh, int flags)
>>>>  {
>>>>  	struct ext4_map_blocks map;
>>>> +	handle_t *handle = ext4_journal_current_handle();
>>>>  	int ret = 0;
>>>>  
>>>>  	if (ext4_has_inline_data(inode))
>>>>  		return -ERANGE;
>>>>  
>>>> +	/* Make sure transaction has enough credits for this extent */
>>>> +	if (flags & EXT4_GET_BLOCKS_CREATE) {
>>>> +		ret = ext4_journal_ensure_extent_credits(handle, inode);
>>>> +		if (ret)
>>>> +			return ret;
>>>> +	}
>>>> +
>>>>  	map.m_lblk = iblock;
>>>>  	map.m_len = bh->b_size >> inode->i_blkbits;
>>>>  
>>>> -	ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
>>>> -			      flags);
>>>> +	ret = ext4_map_blocks(handle, inode, &map, flags);
>>>
>>> Good spotting with ext4_page_mkwrite() and ext4_write_begin() also needing
>>> this treatment! But rather then hiding the transaction extension in
>>> _ext4_get_block() I'd do this in ext4_block_write_begin() where it is much
>>> more obvious (and also it is much more obvious who needs to be prepared for
>>> handling EAGAIN error). Otherwise the patch looks good!
>>>
>>
>> Yes, I completely agree with you. However, unfortunately, do this in
>> ext4_block_write_begin() only works for ext4_write_begin().
>> ext4_page_mkwrite() does not call ext4_block_write_begin() to allocate
>> blocks, it call the vfs helper __block_write_begin_int() instead.
>>
>> vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
>> {
>> 	...
>> 	if (!ext4_should_journal_data(inode)) {
>> 		err = block_page_mkwrite(vma, vmf, get_block);
>> 	...
>> }
>>
>>
>> So...
> 
> Right, I forgot about the nodelalloc case. But since we do most of things
> by hand for data=journal mode, perhaps we could lift some code from
> data=journal mode and reuse it for nodelalloc as well like:
> 
>         folio_lock(folio);
>         size = i_size_read(inode);
>         /* Page got truncated from under us? */
>         if (folio->mapping != mapping || folio_pos(folio) > size) {
>                 ret = VM_FAULT_NOPAGE;
>                 goto out_error;
>         }
> 
>         len = folio_size(folio);
>         if (folio_pos(folio) + len > size)
>                 len = size - folio_pos(folio);
>                 
>         err = ext4_block_write_begin(handle, folio, 0, len,
>                                      get_block);
> 	if (err)
> 		goto out_error;
> 	if (!ext4_should_journal_data(inode))
> 		block_commit_write(folio, 0, len);
> 		folio_mark_dirty(folio);
> 	} else {
> 	        if (ext4_journal_folio_buffers(handle, folio, len)) {
> 	        	ret = VM_FAULT_SIGBUS;
> 		        goto out_error;
> 		}
> 	}
> 	ext4_journal_stop(handle);
> 	folio_wait_stable(folio);
> 
> We get an additional bonus for not waiting for page writeback with
> transaction handle held (which is a potential deadlock vector). What do you
> think?
> 

Yeah, this solution looks nice to me, it should works! Thank you for
the suggestion.

Best regards,
Yi.