Re: [PATCH v2 2/2] xfsprogs: sync transaction log spaces changes to the user space

[Jeff Liu <jeff.liu@oracle.com>]

Hi Mark,


On 01/28/2013 11:57 PM, Mark Tinguely wrote:
> On 01/28/13 07:38, Jeff Liu wrote:
>> Sync the kernel codes changes regarding transaction log space reservations
>> to the user space.
>>
>> As we have splitted the calculation of attrset log space reservations into mount
>> time and runtime in kernel code, here we need to fix max_attrset_trans_res_adjust()
>> to reflect this change.
>>
>> Signed-off-by: Jie Liu<jeff.liu@oracle.com>
>>
>> ---
>>   include/xfs_mount.h |    3 +-
>>   include/xfs_trans.h |    9 +-
>>   libxfs/xfs_attr.c   |    9 +-
>>   libxfs/xfs_trans.c  |  265 ++++++++++++++++++++++++++-------------------------
>>   mkfs/maxtrres.c     |    9 +-
>>   5 files changed, 147 insertions(+), 148 deletions(-)
>>
> 
> For anyone else looking at this. This is basically kernel log 
> reservation patch 2 and 13 combined.
> 
> 
>> diff --git a/include/xfs_trans.h b/include/xfs_trans.h
>> index c2042b7..2870308 100644
>> --- a/include/xfs_trans.h
>> +++ b/include/xfs_trans.h
>> @@ -254,12 +254,9 @@ struct xfs_log_item_desc {
>>   #define	XFS_WRITEID_LOG_RES(mp)	((mp)->m_reservations.tr_swrite)
>>   #define	XFS_ADDAFORK_LOG_RES(mp)	((mp)->m_reservations.tr_addafork)
>>   #define	XFS_ATTRINVAL_LOG_RES(mp)	((mp)->m_reservations.tr_attrinval)
>> -#define	XFS_ATTRSET_LOG_RES(mp, ext)	\
>> -	((mp)->m_reservations.tr_attrset + \
>> -	 (ext * (mp)->m_sb.sb_sectsize) + \
>> -	 (ext * XFS_FSB_TO_B((mp), XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK))) + \
>> -	 (128 * (ext + (ext * XFS_BM_MAXLEVELS(mp, XFS_ATTR_FORK)))))
>> -#define	XFS_ATTRRM_LOG_RES(mp)	((mp)->m_reservations.tr_attrrm)
>> +#define	XFS_ATTRSETM_LOG_RES(mp)	((mp)->m_reservations.tr_attrsetm)
>> +#define	XFS_ATTRSETRT_LOG_RES(mp)	((mp)->m_reservations.tr_attrsetrt)
>> +#define	XFS_ATTRRM_LOG_RES(mp)		((mp)->m_reservations.tr_attrrm)
>>   #define	XFS_CLEAR_AGI_BUCKET_LOG_RES(mp)  ((mp)->m_reservations.tr_clearagi)
> 
>   ...
> 
>> index 33dee52..bf8a2c5 100644
>> --- a/mkfs/maxtrres.c
>> +++ b/mkfs/maxtrres.c
>> @@ -34,7 +34,6 @@ max_attrset_trans_res_adjust(
>>   	int				local;
>>   	int				size;
>>   	int				nblks;
>> -	int				res;
>>
>>   	/*
>>   	 * Determine space the maximal sized attribute will use,
>> @@ -46,13 +45,9 @@ max_attrset_trans_res_adjust(
>>   	nblks = XFS_DAENTER_SPACE_RES(mp, XFS_ATTR_FORK);
>>   	nblks += XFS_B_TO_FSB(mp, size);
>>   	nblks += XFS_NEXTENTADD_SPACE_RES(mp, size, XFS_ATTR_FORK);
>> -	res = XFS_ATTRSET_LOG_RES(mp, nblks);
>>
>> -#if 0
>> -	printf("size = %d nblks = %d res = %d\n", size, nblks, res);
>> -#endif
>> -
>> -	mp->m_reservations.tr_attrset = res;
> 
>    This number used to be the mount portion of constant that was
>    multiplied times args.total.
> 
>    Isn't this value (res) what we want to place into tr_attrsetm
>    like before?
This is not because we have to figure out the max transaction space
reservations for the attrset at first, and then it will be used for
the max trans space reservations by mount...

The old attrset includes both mount(attrsetm) and realtime(attrsetrt), that's
why I splitted them as below.
> 
>    and then tr_attrsetrt is from xfs_calc_attrsetrt_reservation()?
> 
>> +	mp->m_reservations.tr_attrsetm = XFS_ATTRSETM_LOG_RES(mp);
>> +	mp->m_reservations.tr_attrsetrt = XFS_ATTRSETRT_LOG_RES(mp) * nblks;
>>   }
>>
I have tested it by putting printf statement to max_attrset_trans_res_adjust(),
and verified that the "res = XFS_ATTRSET_LOG_RES(mp, nblks)" is equal to the
"XFS_ATTRSETM_LOG_RES(mp) + XFS_ATTRSETRT_LOG_RES(mp) * nblks".

Thanks,
-Jeff

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs