[PATCH] xfs: perag initialization should only touch m_ag_max_usable for AG 0

[PATCH] xfs: perag initialization should only touch m_ag_max_usable for AG 0

[Darrick J. Wong]

We call __xfs_ag_resv_init to make a per-AG reservation for each AG.
This makes the reservation per-AG, not per-filesystem.  Therefore, it
is incorrect to adjust m_ag_max_usable for each AG.  Adjust it only
when we're reserving AG 0's blocks so that we only do it once per fs.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
---
 fs/xfs/libxfs/xfs_ag_resv.c |   12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/fs/xfs/libxfs/xfs_ag_resv.c b/fs/xfs/libxfs/xfs_ag_resv.c
index b008ff3..df3e600 100644
--- a/fs/xfs/libxfs/xfs_ag_resv.c
+++ b/fs/xfs/libxfs/xfs_ag_resv.c
@@ -156,7 +156,8 @@ __xfs_ag_resv_free(
 	trace_xfs_ag_resv_free(pag, type, 0);
 
 	resv = xfs_perag_resv(pag, type);
-	pag->pag_mount->m_ag_max_usable += resv->ar_asked;
+	if (pag->pag_agno == 0)
+		pag->pag_mount->m_ag_max_usable += resv->ar_asked;
 	/*
 	 * AGFL blocks are always considered "free", so whatever
 	 * was reserved at mount time must be given back at umount.
@@ -216,7 +217,14 @@ __xfs_ag_resv_init(
 		return error;
 	}
 
-	mp->m_ag_max_usable -= ask;
+	/*
+	 * Reduce the maximum per-AG allocation length by however much we're
+	 * trying to reserve for an AG.  Since this is a filesystem-wide
+	 * counter, we only make the adjustment for AG 0.  This assumes that
+	 * there aren't any AGs hungrier for per-AG reservation than AG 0.
+	 */
+	if (pag->pag_agno == 0)
+		mp->m_ag_max_usable -= ask;
 
 	resv = xfs_perag_resv(pag, type);
 	resv->ar_asked = ask;

Re: [PATCH] xfs: perag initialization should only touch m_ag_max_usable for AG 0

[Brian Foster]

On Tue, Sep 19, 2017 at 11:10:41AM -0700, Darrick J. Wong wrote:
> We call __xfs_ag_resv_init to make a per-AG reservation for each AG.
> This makes the reservation per-AG, not per-filesystem.  Therefore, it
> is incorrect to adjust m_ag_max_usable for each AG.  Adjust it only
> when we're reserving AG 0's blocks so that we only do it once per fs.
> 
> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> ---

Reviewed-by: Brian Foster <bfoster@redhat.com>

>  fs/xfs/libxfs/xfs_ag_resv.c |   12 ++++++++++--
>  1 file changed, 10 insertions(+), 2 deletions(-)
> 
> diff --git a/fs/xfs/libxfs/xfs_ag_resv.c b/fs/xfs/libxfs/xfs_ag_resv.c
> index b008ff3..df3e600 100644
> --- a/fs/xfs/libxfs/xfs_ag_resv.c
> +++ b/fs/xfs/libxfs/xfs_ag_resv.c
> @@ -156,7 +156,8 @@ __xfs_ag_resv_free(
>  	trace_xfs_ag_resv_free(pag, type, 0);
>  
>  	resv = xfs_perag_resv(pag, type);
> -	pag->pag_mount->m_ag_max_usable += resv->ar_asked;
> +	if (pag->pag_agno == 0)
> +		pag->pag_mount->m_ag_max_usable += resv->ar_asked;
>  	/*
>  	 * AGFL blocks are always considered "free", so whatever
>  	 * was reserved at mount time must be given back at umount.
> @@ -216,7 +217,14 @@ __xfs_ag_resv_init(
>  		return error;
>  	}
>  
> -	mp->m_ag_max_usable -= ask;
> +	/*
> +	 * Reduce the maximum per-AG allocation length by however much we're
> +	 * trying to reserve for an AG.  Since this is a filesystem-wide
> +	 * counter, we only make the adjustment for AG 0.  This assumes that
> +	 * there aren't any AGs hungrier for per-AG reservation than AG 0.
> +	 */
> +	if (pag->pag_agno == 0)
> +		mp->m_ag_max_usable -= ask;
>  
>  	resv = xfs_perag_resv(pag, type);
>  	resv->ar_asked = ask;
> --
> To unsubscribe from this list: send the line "unsubscribe linux-xfs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH] xfs: perag initialization should only touch m_ag_max_usable for AG 0

[Christoph Hellwig]

On Tue, Sep 19, 2017 at 11:10:41AM -0700, Darrick J. Wong wrote:
> We call __xfs_ag_resv_init to make a per-AG reservation for each AG.
> This makes the reservation per-AG, not per-filesystem.  Therefore, it
> is incorrect to adjust m_ag_max_usable for each AG.  Adjust it only
> when we're reserving AG 0's blocks so that we only do it once per fs.

This looks correct to me, but I hate how we manipulate a per-mount
value.  Is there a way we can just move this to the caller?

Re: [PATCH] xfs: perag initialization should only touch m_ag_max_usable for AG 0

[Darrick J. Wong]

On Thu, Sep 21, 2017 at 07:04:21AM -0700, Christoph Hellwig wrote:
> On Tue, Sep 19, 2017 at 11:10:41AM -0700, Darrick J. Wong wrote:
> > We call __xfs_ag_resv_init to make a per-AG reservation for each AG.
> > This makes the reservation per-AG, not per-filesystem.  Therefore, it
> > is incorrect to adjust m_ag_max_usable for each AG.  Adjust it only
> > when we're reserving AG 0's blocks so that we only do it once per fs.
> 
> This looks correct to me, but I hate how we manipulate a per-mount
> value.  Is there a way we can just move this to the caller?

I'm not sure which caller you're referring to -- if you mean
xfs_ag_resv_init, then we'd still have the "if (pag->agno == 0)
xfs_mod_fdblocks(...)"; and if xfs_fs_reserve_ag_blocks then you'd have
to bubble the two ask values up out of xfs_ag_resv_init, which also
seems like a layering violation.  Modifying fdblocks in
__xfs_ag_resv_init (Dave I think is sharpening his axe on that) smells
kinda funny but it's less code intensive.

<confused>

--D

> --
> To unsubscribe from this list: send the line "unsubscribe linux-xfs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html