We find that xfs_inode_set_reclaim_tag() and xfs_reclaim_inode() are
concurrently executed at runtime in the following call contexts:

Thread1:
  xfs_fs_put_super()
    xfs_unmountfs()
      xfs_rtunmount_inodes()
        xfs_irele()
          xfs_fs_destroy_inode()
            xfs_inode_set_reclaim_tag()

Thread2:
  xfs_reclaim_worker()
    xfs_reclaim_inodes()
      xfs_reclaim_inodes_ag()
        xfs_reclaim_inode()

In xfs_inode_set_reclaim_tag():
  pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
  ...
  spin_lock(&ip->i_flags_lock);

In xfs_reclaim_inode():
  spin_lock(&ip->i_flags_lock);
  ...
  ip->i_ino = 0;
  spin_unlock(&ip->i_flags_lock);

Thus, a data race can occur for ip->i_ino.

To fix this data race, the spinlock ip->i_flags_lock is used to protect
the access to ip->i_ino in xfs_inode_set_reclaim_tag().

This data race is found by our concurrency fuzzer.

Signed-off-by: Jia-Ju Bai <[email protected]>
---
 fs/xfs/xfs_icache.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/xfs/xfs_icache.c b/fs/xfs/xfs_icache.c
index 8bf1d15be3f6..a2de08222ff5 100644
--- a/fs/xfs/xfs_icache.c
+++ b/fs/xfs/xfs_icache.c
@@ -229,9 +229,9 @@ xfs_inode_set_reclaim_tag(
        struct xfs_mount        *mp = ip->i_mount;
        struct xfs_perag        *pag;
 
+       spin_lock(&ip->i_flags_lock);
        pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
        spin_lock(&pag->pag_ici_lock);
-       spin_lock(&ip->i_flags_lock);
 
        radix_tree_tag_set(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, ip->i_ino),
                           XFS_ICI_RECLAIM_TAG);
-- 
2.17.1

Reply via email to