Re: [PATCH v2 3/4] xfs: check tp->t_dqinfo value instead of the XFS_TRANS_DQ_DIRTY flag

[kaixuxia <xiakaixu1987@gmail.com>]

On 2020/9/26 23:36, Eric Sandeen wrote:
> On 9/26/20 8:14 AM, xiakaixu1987@gmail.com wrote:
>> From: Kaixu Xia <kaixuxia@tencent.com>
>>
>> Nowadays the only things that the XFS_TRANS_DQ_DIRTY flag seems to do
>> are indicates the tp->t_dqinfo->dqs[XFS_QM_TRANS_{USR,GRP,PRJ}] values
>> changed and check in xfs_trans_apply_dquot_deltas() and the unreserve
>> variant xfs_trans_unreserve_and_mod_dquots(). Actually, we also can
>> use the tp->t_dqinfo value instead of the XFS_TRANS_DQ_DIRTY flag, that
>> is to say, we allocate the new tp->t_dqinfo only when the qtrx values
>> changed, so the tp->t_dqinfo value isn't NULL equals the XFS_TRANS_DQ_DIRTY
>> flag is set, we only need to check if tp->t_dqinfo == NULL in
>> xfs_trans_apply_dquot_deltas() and its unreserve variant to determine
>> whether lock all of the dquots and join them to the transaction.
>>
>> Signed-off-by: Kaixu Xia <kaixuxia@tencent.com>
>> ---
>>  fs/xfs/libxfs/xfs_shared.h |  1 -
>>  fs/xfs/xfs_inode.c         |  8 +-------
>>  fs/xfs/xfs_trans_dquot.c   | 20 ++------------------
>>  3 files changed, 3 insertions(+), 26 deletions(-)
>>
>> diff --git a/fs/xfs/libxfs/xfs_shared.h b/fs/xfs/libxfs/xfs_shared.h
>> index c795ae47b3c9..8c61a461bf7b 100644
>> --- a/fs/xfs/libxfs/xfs_shared.h
>> +++ b/fs/xfs/libxfs/xfs_shared.h
>> @@ -62,7 +62,6 @@ void	xfs_log_get_max_trans_res(struct xfs_mount *mp,
>>  #define	XFS_TRANS_SB_DIRTY	0x02	/* superblock is modified */
>>  #define	XFS_TRANS_PERM_LOG_RES	0x04	/* xact took a permanent log res */
>>  #define	XFS_TRANS_SYNC		0x08	/* make commit synchronous */
>> -#define XFS_TRANS_DQ_DIRTY	0x10	/* at least one dquot in trx dirty */
>>  #define XFS_TRANS_RESERVE	0x20    /* OK to use reserved data blocks */
>>  #define XFS_TRANS_NO_WRITECOUNT 0x40	/* do not elevate SB writecount */
>>  #define XFS_TRANS_RES_FDBLKS	0x80	/* reserve newly freed blocks */
>> diff --git a/fs/xfs/xfs_inode.c b/fs/xfs/xfs_inode.c
>> index 49624973eecc..9108eed0ea45 100644
>> --- a/fs/xfs/xfs_inode.c
>> +++ b/fs/xfs/xfs_inode.c
>> @@ -941,7 +941,6 @@ xfs_dir_ialloc(
>>  	xfs_buf_t	*ialloc_context = NULL;
>>  	int		code;
>>  	void		*dqinfo;
>> -	uint		tflags;
>>  
>>  	tp = *tpp;
>>  	ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
>> @@ -1000,12 +999,9 @@ xfs_dir_ialloc(
>>  		 * and attach it to the next transaction.
>>  		 */
>>  		dqinfo = NULL;
>> -		tflags = 0;
>>  		if (tp->t_dqinfo) {
>>  			dqinfo = (void *)tp->t_dqinfo;
>>  			tp->t_dqinfo = NULL;
>> -			tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY;
>> -			tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY);
>>  		}
>>  
>>  		code = xfs_trans_roll(&tp);
>> @@ -1013,10 +1009,8 @@ xfs_dir_ialloc(
>>  		/*
>>  		 * Re-attach the quota info that we detached from prev trx.
>>  		 */
>> -		if (dqinfo) {
>> +		if (dqinfo)
>>  			tp->t_dqinfo = dqinfo;
>> -			tp->t_flags |= tflags;
>> -		}
>>  
>>  		if (code) {
>>  			xfs_buf_relse(ialloc_context);
>> diff --git a/fs/xfs/xfs_trans_dquot.c b/fs/xfs/xfs_trans_dquot.c
>> index fe45b0c3970c..0ebfd7930382 100644
>> --- a/fs/xfs/xfs_trans_dquot.c
>> +++ b/fs/xfs/xfs_trans_dquot.c
>> @@ -84,13 +84,6 @@ xfs_trans_dup_dqinfo(
>>  
>>  	xfs_trans_alloc_dqinfo(ntp);
>>  
>> -	/*
>> -	 * Because the quota blk reservation is carried forward,
>> -	 * it is also necessary to carry forward the DQ_DIRTY flag.
>> -	 */
>> -	if (otp->t_flags & XFS_TRANS_DQ_DIRTY)
>> -		ntp->t_flags |= XFS_TRANS_DQ_DIRTY;
>> -
>>  	for (j = 0; j < XFS_QM_TRANS_DQTYPES; j++) {
>>  		oqa = otp->t_dqinfo->dqs[j];
>>  		nqa = ntp->t_dqinfo->dqs[j];
>> @@ -143,9 +136,6 @@ xfs_trans_mod_dquot_byino(
>>  	    xfs_is_quota_inode(&mp->m_sb, ip->i_ino))
>>  		return;
>>  
>> -	if (tp->t_dqinfo == NULL)
>> -		xfs_trans_alloc_dqinfo(tp);
>> -
> 
> I can't tell from the commit log or from a very quick read of the code why these
> allocations are being removed.  Can we not get here with a NULL t_dqinfo?
> If not, why not?  This seems like a change unrelated to the proposed
> "t_dqinfo set == XFS_TRANS_DQ_DIRTY" change.

Yeah, remove these allocations because I want to allocate the t_dqinfo only
when the tp->t_dqinfo->dqs[XFS_QM_TRANS_{USR,GRP,PRJ}] values changed, that
is to say, only do the allocation in xfs_trans_mod_dquot() function. Actually,
original these allocations are repeated, for example, the xfs_trans_mod_dquot_byino()
function call the xfs_trans_mod_dquot(), but both of them do the allocation,
so remove one of them may be reasonable.  
> 
> Also, while it seems clear to say that !t_dqinfo == !XFS_TRANS_DQ_DIRTY, is the
> converse true?  Is it possible to have t_dqinfo set, but it's not dirty?> 
> I think the answer is that when we free the transaction we set t_dqinfo to
> NULL again, but I'm not certain, and it's not obvious from the changelog...

Now we do the allocation in xfs_trans_mod_dquot() function and only have t_dqinfo
set when it is dirty.

Thanks,
Kaixu
> 
> -Eric
> 

-- 
kaixuxia