[PATCH v2 02/12] xfs: make use of xfs_calc_buf_res() in xfs_trans.c

[Mark Tinguely]

On 01/19/13 17:09, Dave Chinner wrote:
> On Sat, Jan 19, 2013 at 01:08:47PM -0600, Mark Tinguely wrote:
>> On 01/10/13 07:47, Jeff Liu wrote:
>>> Start to make use of the new helper to figure out space log reservations for those
>>> transactions which are pre-calculated at mount time in xfs_trans.c.
>>>
>>> Signed-off-by: Jie Liu<jeff.liu at oracle.com>
>>> ---
>>>   fs/xfs/xfs_trans.c |  244 ++++++++++++++++++++++++----------------------------
>>>   1 file changed, 113 insertions(+), 131 deletions(-)
>>>
>>
>> Wow! Reading this patch makes me appreciate the work you did here
>> and gets my eyes in shape for Dave's UBER user sync patch.
>>
>> A question for you, or anyone. When these reservations are made, the
>> comments talk about specify number of agf/agfl (usually 2 or 3) that
>> will be dirty in the command.
>>
>> There are other comments that seem to imply an agf/agfl is reserved
>> for all AGs and then use the multiplier of 4. Is a specific number
>> of AGs can be involved in the operation or does it want something
>> like sb_agcount?
>
> Do you mean comments like this about
> xfs_calc_itruncate_reservation()?
>
>   * And the bmap_finish transaction can free the blocks and bmap
>   * blocks:
>   *    the agf for each of the ags: 4 * sector size
>   *    the agfl for each of the ags: 4 * sector size
>
> This assumes the transaction can free 4 extents before a commit, and
> all 4 extents can be in a different AG.
>
> You'll find all the other cases documented like this indicate how
> many extents can be freed or allocated in a single transaction....
>


Yes thanks.

The 2/3 cases were obvious, but I was wondering how to read this one.

--Mark.