xfs: account for the refcount btree in the alloc/free log reservation

Every time we allocate or free a data extent, we might need to split
the refcount btree.  Reserve some blocks in the transaction to handle
this possibility.  Even though the deferred refcount code can roll a
transaction to avoid overloading the transaction, we can still exceed
the reservation.

Certain pathological workloads (1k blocks, no cowextsize hint, random
directio writes), cause a perfect storm wherein a refcount adjustment
of a large range of blocks causes full tree splits in two separate
extents in two separate refcount tree blocks; allocating new refcount
tree blocks causes rmap btree splits; and all the allocation activity
causes the freespace btrees to split, blowing the reservation.

(Reproduced by generic/167 over NFS atop XFS)

Signed-off-by: Christoph Hellwig <hch@lst.de>
[darrick.wong@oracle.com: add commit message]
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>

diff --git a/fs/xfs/libxfs/xfs_trans_resv.c b/fs/xfs/libxfs/xfs_trans_resv.c
index 7c840e1..a59838f 100644 (file)
--- a/fs/xfs/libxfs/xfs_trans_resv.c
+++ b/fs/xfs/libxfs/xfs_trans_resv.c
@@ -67,7 +67,8 @@ xfs_calc_buf_res(
  * Per-extent log reservation for the btree changes involved in freeing or
  * allocating an extent.  In classic XFS there were two trees that will be
  * modified (bnobt + cntbt).  With rmap enabled, there are three trees
- * (rmapbt).  The number of blocks reserved is based on the formula:
+ * (rmapbt).  With reflink, there are four trees (refcountbt).  The number of
+ * blocks reserved is based on the formula:
  *
  * num trees * ((2 blocks/level * max depth) - 1)
  *
@@ -83,6 +84,8 @@ xfs_allocfree_log_count(
        blocks = num_ops * 2 * (2 * mp->m_ag_maxlevels - 1);
        if (xfs_sb_version_hasrmapbt(&mp->m_sb))
                blocks += num_ops * (2 * mp->m_rmap_maxlevels - 1);
+       if (xfs_sb_version_hasreflink(&mp->m_sb))
+               blocks += num_ops * (2 * mp->m_refc_maxlevels - 1);
 
        return blocks;
 }