swapon: Document btrfs limitation with swapfiles

Btrfs, as of 2.6.35, is unable to allow swapfiles to be used on its
filesystems. This is due to the swapfile implementation wanting to
build an extent map of each block in the file and expecting it to be
static for the life of the swapfile.

Btrfs can't guarantee this and refuses to return the mapping. The
swapfile implementation just makes a comment about there being holes
in the file - but that's how btrfs denies the mapping.

This patch adds a section to the swapon manpage to document it.

Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Karel Zak <kzak@redhat.com>

diff --git a/mount/swapon.8 b/mount/swapon.8
index 25fc5e81def35abd91de2eb1502799a34783fd7c..a3ebc97cdaa8fb08058117d887fbe897bf79b309 100644 (file)
--- a/mount/swapon.8
+++ b/mount/swapon.8
@@ -167,6 +167,18 @@ automatically detects and rewrites swap space signature with old software
 suspend data (e.g S1SUSPEND, S2SUSPEND, ...). The problem is that if we don't
 do it, then we get data corruption the next time an attempt at unsuspending is
 made.
+.PP
+.B swapon
+may not work correctly when using a swap file with some versions of btrfs.
+This is due to the swap file implementation in the kernel expecting to be able
+to write to the file directly, without the assistance of the file system.
+Since btrfs is a copy-on-write file system, the file location may not be
+static and corruption can result. Btrfs actively disallows the use of files
+on its file systems by refusing to map the file. This can be seen in the system
+log as "swapon: swapfile has holes." One possible workaround is to map the
+file to a loopback device. This will allow the file system to determine the
+mapping properly but may come with a performance impact.
+
 .SH SEE ALSO
 .BR swapon (2),
 .BR swapoff (2),