Fix the brand new xfstest that tries to swapon on a recently unshared
file and use the chance to document the other bit of magic in this
function.
The big comment is taken from a mailinglist post by Dave Chinner.
Fixes: 5e672cd69f0a53 ("xfs: introduce xfs_inodegc_push()")
Signed-off-by: Christoph Hellwig <hch@xxxxxx>
Reviewed-by: "Darrick J. Wong" <djwong@xxxxxxxxxx>
---
fs/xfs/xfs_aops.c | 35 ++++++++++++++++++++++++++++++++++-
1 file changed, 34 insertions(+), 1 deletion(-)
diff --git a/fs/xfs/xfs_aops.c b/fs/xfs/xfs_aops.c
index 69b8c2d1937d..9ff1a0b07303 100644
--- a/fs/xfs/xfs_aops.c
+++ b/fs/xfs/xfs_aops.c
@@ -21,6 +21,7 @@
#include "xfs_error.h"
#include "xfs_zone_alloc.h"
#include "xfs_rtgroup.h"
+#include "xfs_icache.h"
struct xfs_writepage_ctx {
struct iomap_writepage_ctx ctx;
@@ -685,7 +686,39 @@ xfs_iomap_swapfile_activate(
struct file *swap_file,
sector_t *span)
{
- sis->bdev = xfs_inode_buftarg(XFS_I(file_inode(swap_file)))->bt_bdev;
+ struct xfs_inode *ip = XFS_I(file_inode(swap_file));
+
+ /*
+ * Swap file activation can race against concurrent shared extent
+ * removal in files that have been cloned. If this happens,
+ * iomap_swapfile_iter() can fail because it encountered a shared
+ * extent even though an operation is in progress to remove those
+ * shared extents.
+ *
+ * This race becomes problematic when we defer extent removal
+ * operations beyond the end of a syscall (i.e. use async background
+ * processing algorithms). Users think the extents are no longer
+ * shared, but iomap_swapfile_iter() still sees them as shared
+ * because the refcountbt entries for the extents being removed have
+ * not yet been updated. Hence the swapon call fails unexpectedly.
+ *
+ * The race condition is currently most obvious from the unlink()
+ * operation as extent removal is deferred until after the last
+ * reference to the inode goes away. We then process the extent
+ * removal asynchronously, hence triggers the "syscall completed but
+ * work not done" condition mentioned above. To close this race
+ * window, we need to flush any pending inodegc operations to ensure
+ * they have updated the refcountbt records before we try to map the
+ * swapfile.
+ */
+ xfs_inodegc_flush(ip->i_mount);
+
+ /*
+ * Direct the swap code to the correct block device when this file
+ * sits on the RT device.
+ */
+ sis->bdev = xfs_inode_buftarg(ip)->bt_bdev;
+
return iomap_swapfile_activate(sis, swap_file, span,
&xfs_read_iomap_ops);
}
--
2.45.2
