xfs: reserve blocks for ifree transaction during log recovery

Log recovery frees all the inodes stored in the unlinked list, which can
cause expansion of the free inode btree.  The ifree code skips block
reservations if it thinks there's a per-AG space reservation, but we
don't set up the reservation until after log recovery, which means that
a finobt expansion blows up in xfs_trans_mod_sb when we exceed the
transaction's block reservation.

To fix this, we set the "no finobt reservation" flag to true when we
create the xfs_mount and only set it to false if we confirm that every
AG had enough free space to put aside for the finobt.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>

diff --git a/fs/xfs/xfs_fsops.c b/fs/xfs/xfs_fsops.c
index f3ef70c542e1bc3392b391fefe904b787a0920a5..584648582ba717be54e61f7829085025acbc2ea3 100644 (file)
--- a/fs/xfs/xfs_fsops.c
+++ b/fs/xfs/xfs_fsops.c
@@ -533,6 +533,7 @@ xfs_fs_reserve_ag_blocks(
        int                     error = 0;
        int                     err2;
 
+       mp->m_finobt_nores = false;
        for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
                pag = xfs_perag_get(mp, agno);
                err2 = xfs_ag_resv_init(pag, NULL);

diff --git a/fs/xfs/xfs_super.c b/fs/xfs/xfs_super.c
index c9097cb0b955ea9219185a97551808befc7da495..08033ac040d6cf743e23536cd0484b06ce29c358 100644 (file)
--- a/fs/xfs/xfs_super.c
+++ b/fs/xfs/xfs_super.c
@@ -1594,6 +1594,13 @@ xfs_mount_alloc(
        INIT_DELAYED_WORK(&mp->m_eofblocks_work, xfs_eofblocks_worker);
        INIT_DELAYED_WORK(&mp->m_cowblocks_work, xfs_cowblocks_worker);
        mp->m_kobj.kobject.kset = xfs_kset;
+       /*
+        * We don't create the finobt per-ag space reservation until after log
+        * recovery, so we must set this to true so that an ifree transaction
+        * started during log recovery will not depend on space reservations
+        * for finobt expansion.
+        */
+       mp->m_finobt_nores = true;
        return mp;
 }