Re: [PATCH 5/5] xfs: avoid unnecessary waits in xfs_log_force_lsn()

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Thu, Jul 22, 2021 at 11:53:35AM +1000, Dave Chinner wrote:
> From: Dave Chinner <dchinner@xxxxxxxxxx>
> 
> Before waiting on a iclog in xfs_log_force_lsn(), we don't check to
> see if the iclog has already been completed and the contents on
> stable storage. We check for completed iclogs in xfs_log_force(), so
> we should do the same thing for xfs_log_force_lsn().
> 
> This fixed some random up-to-30s pauses seen in unmounting
> filesystems in some tests. A log force ends up waiting on completed
> iclog, and that doesn't then get flushed (and hence the log force
> get completed) until the background log worker issues a log force
> that flushes the iclog in question. Then the unmount unblocks and
> continues.
> 
> Signed-off-by: Dave Chinner <dchinner@xxxxxxxxxx>

Heh.
Reviewed-by: Darrick J. Wong <djwong@xxxxxxxxxx>

--D

> ---
>  fs/xfs/xfs_log.c | 47 +++++++++++++++++++++++++++++++++++++----------
>  1 file changed, 37 insertions(+), 10 deletions(-)
> 
> diff --git a/fs/xfs/xfs_log.c b/fs/xfs/xfs_log.c
> index 7da42c0656e3..baee9871cd65 100644
> --- a/fs/xfs/xfs_log.c
> +++ b/fs/xfs/xfs_log.c
> @@ -3147,6 +3147,35 @@ xlog_state_switch_iclogs(
>  	log->l_iclog = iclog->ic_next;
>  }
>  
> +/*
> + * Force the iclog to disk and check if the iclog has been completed before
> + * xlog_force_iclog() returns. This can happen on synchronous (e.g.
> + * pmem) or fast async storage because we drop the icloglock to issue the IO.
> + * If completion has already occurred, tell the caller so that it can avoid an
> + * unnecessary wait on the iclog.
> + */
> +static int
> +xlog_force_and_check_iclog(
> +	struct xlog_in_core	*iclog,
> +	bool			*completed)
> +{
> +	xfs_lsn_t		lsn = be64_to_cpu(iclog->ic_header.h_lsn);
> +	int			error;
> +
> +	*completed = false;
> +	error = xlog_force_iclog(iclog);
> +	if (error)
> +		return error;
> +
> +	/*
> +	 * If the iclog has already been completed and reused the header LSN
> +	 * will have been rewritten by completion
> +	 */
> +	if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn)
> +		*completed = true;
> +	return 0;
> +}
> +
>  /*
>   * Write out all data in the in-core log as of this exact moment in time.
>   *
> @@ -3181,7 +3210,6 @@ xfs_log_force(
>  {
>  	struct xlog		*log = mp->m_log;
>  	struct xlog_in_core	*iclog;
> -	xfs_lsn_t		lsn;
>  
>  	XFS_STATS_INC(mp, xs_log_force);
>  	trace_xfs_log_force(mp, 0, _RET_IP_);
> @@ -3210,15 +3238,11 @@ xfs_log_force(
>  	} else if (iclog->ic_state == XLOG_STATE_ACTIVE) {
>  		if (atomic_read(&iclog->ic_refcnt) == 0) {
>  			/* We have exclusive access to this iclog. */
> -			lsn = be64_to_cpu(iclog->ic_header.h_lsn);
> -			if (xlog_force_iclog(iclog))
> +			bool	completed;
> +
> +			if (xlog_force_and_check_iclog(iclog, &completed))
>  				goto out_error;
> -			/*
> -			 * If the iclog has already been completed and reused
> -			 * the header LSN will have been rewritten. Don't wait
> -			 * on these iclogs that contain future modifications.
> -			 */
> -			if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn)
> +			if (completed)
>  				goto out_unlock;
>  		} else {
>  			/*
> @@ -3258,6 +3282,7 @@ xlog_force_lsn(
>  	bool			already_slept)
>  {
>  	struct xlog_in_core	*iclog;
> +	bool			completed;
>  
>  	spin_lock(&log->l_icloglock);
>  	iclog = log->l_iclog;
> @@ -3295,10 +3320,12 @@ xlog_force_lsn(
>  					&log->l_icloglock);
>  			return -EAGAIN;
>  		}
> -		if (xlog_force_iclog(iclog))
> +		if (xlog_force_and_check_iclog(iclog, &completed))
>  			goto out_error;
>  		if (log_flushed)
>  			*log_flushed = 1;
> +		if (completed)
> +			goto out_unlock;
>  		break;
>  	case XLOG_STATE_WANT_SYNC:
>  		/*
> -- 
> 2.31.1
> 



[Index of Archives]     [XFS Filesystem Development (older mail)]     [Linux Filesystem Development]     [Linux Audio Users]     [Yosemite Trails]     [Linux Kernel]     [Linux RAID]     [Linux SCSI]


  Powered by Linux